In Photoshop, you create a custom RGB setting with a gamma of 1.0. Edit>Convert to Profile. Click on the Profile drop down menu and select Custom RGB. Then type in a gamma of 1.0. You may also want to change the Primaries to Adobe RGB 1998 (mine defaults to HDTV for some reason). As for any other programs, couldn't tell ya.
I'm a professional illustrator, and you just taught me something. I'd always assumed it was because image editors were intentionally treating colours like pigments and mixing them subtractively instead of additively, since the result generally fits.
Interestingly, the non-linear storage of brightness in computer graphics did not evolve as a clever deliberate choice; instead, it was merely a legacy from the display systems used back then: Cathode ray tubes. Their brightness happens to be roughly proportional to the square of the control voltage. Designers of TV broadcasting norms were aware of this, and decided to compensate for this effect in the broadcasting side of the system, to keep the receivers as simple as possible. When those same receivers were later adapted as computer displays, the computer engineers never seemed to have paid any attention to this detail. It was only when computers started to be used in the printing industry that this quirk started to get any attention in computer technology.
@@brod515 Roughly speaking, taking a square-root means keeping the first half of a numbers most significant bits and dropping the other half, essentially cutting the size in half. This is more complicated in practics, but I how you get the idea.
@@Ruhrpottpatriot This still doesn't quite make sense. information like that would be stored in 4 byte floating point numbers which will still use all the bits to represent a number. I don't think that's what he was referring to as compression... there is a common idea that storing the values non-linearly stores only the useful information for the human eye and I don't quite understand it.
As a computer programmer, I think this is less an issue of laziness and more an issue of not realizing the color values were square-rooted in the first place. Thanks for sharing this information.
I noticed how as an artist I never use even a tenth of all the very bright white values available to me, and it irritates me a lot when my dark grey gets 1 unit closer to black but it looks much darker...
Gamma correction is like daylight savings time. The actual mathematical operation is super easy, but I can't for the life of me figure out whether to do the one step or its inverse. I keep rewatching this video time and again.
Regnor Math isn't a science (there's nothing empirical about it) What is meant by that quote is that at the time physics was a reductionist use of math make testable predictions, most other sciences were still concerned with just phenomenology.
Abraham Animations Does it just fine, if you toggle the right mode. The fact that barely anyone noticed this thing prior to this video just speaks volumes how little it matters. And for those whom it does matter (See graphic designers, Texture artists, etc) Photoshop has the option right there for them already, even back in the earlier incarnations of the software, because they know who uses it.
As a student of computer science I can say this is accurate. We learn to blur images with the wrong approach and then with the good approach. It's about understanding how computer graphics work, the same for bubble sort, we learn the easiest method first. What is wrong is have the wrong method in professional tool as the video says.
+Sebb747 like the video says our human vision can't tell the difference between bright colors but dark colors. So, instead of wasting data storing bright colors we can have a better image by storing the root of the original bright value. It's like the mp3 format, instead of saving inaudible sounds we delete those frequencies.
Angel Alvarado Yes, I'm well aware of this. I'm in CS myself. But if you do image processing, you usually convert your image into an RGB(A) array which you then work with instead of working with the raw data of whatever image the user chose to supply to you. My Question was whether those RGB values are representing square roots and are being multiplied down the graphics pipeline or whether this is just a problem for people who choose to - for whatever reason - work with the raw image data.
+Sebb747 You can't know unless you have the data from the original source, take for instance a camera, you can set the gamma values on it but once the photo/video is taken all is stored in the basic RGB(A) values. The same when displaying the image, you can change the gamma values in your TV or screen. The thing for us as developers is how to treat those pixels, you can choose the lazy path and use the mean to "blur" the image or be aware that it's not that simple and you need to consider all cases. Color math is an interesting topic as well. I stopped learning about IP but there are a lot of resources out there.
performance. computers haven't always been this fast. the image formats come from a time where desktop computers were slower than your average smart phone. he was even making this point in the video.
***** That was in the past... however in software in the present still use the same technique... answer... laziness... trust me I'm first year in IT engineering...
VirtualCoder Except they already exist and nobody uses them.The same way there has been an alternative to .docx that's one trillion times better and nobody uses it.And the same way averyone should be using .webm instead of .gif but then again nobodo does.
As a (hobby) programmer, I come back to this video occasionally to remind myself about this. Thank you. On this watch I realized I've programmed contrast wrong in a few programs.
More than one million views and no comments?? As a person who does does a lot of graphic art on the computer, I'm amazed I didn't know this before. And why this has not been fixed.
+Mason Bially In general, the downsides of open software with regards to bugginess, lack of support, and most importantly, lack of economic incentive to fix problems, far outweigh any theoretical advantages. _"Also, you as the user can always fix the problems with open software."_ Absolutely not. Almost no users are competent to fix problems with any large software package - even if they happen to be professional programmers - that stuff is just a pipe dream in 99% of cases, possibly slightly better if the problem can be fixed by writing a plugin. The most laughable part of it though, is that if I, as a programmer, ran into a problem with an open source image processing package and knew what the problem was, it would almost certainly be 100 times quicker for me to write my own special purpose program to just do that particular job instead of spending days or weeks attempting to understand the original program to the point where I could modify it safely without breaking other things. The typical case is that open source software is written by a handful of dedicated enthusiasts, with minimal programming input from users. Blender is a prime example of this, as large critical portions of the program is developed by one guy.(I know this because the lack of development effort prevented me from using the software at one point and it was decidedly not worth my time to write the software myself when I could just buy it from an actual business)
The reason it hasn't been fixed is for larger images, the ammount of time a proper blur takes is far more meaningful then smaller images. a 1024x1024 image has just over 1 million pixels. The first method uses an addition and division operation per pixel (The colors are already square rooted), So for the picture using the first method, just over 1 million additions and divisions The second method uses 1 addition, 1 division, 2 multiplications (squares), and 1 square root (The most expensive basic math function a computer can do(Not counting trig functions)). So this multiplied by 1 million, and it would take 1 million additions, divisions, and square roots, and 2 million multiplications. If you are going for a faster program with less wait time between blurring operations (paint, photoshop) or less intense software on older hardware in general, you go with the 2 million operations rather then the 5 million operations
You're right, like i'm totaly sure the apple engineers just accidently put 2gb or DDR2 RAM in a computer that has 2 4GHz quad core processors. It totaly wasn't just to scam idiots out of their money or anything.
Damb I remember when this video was new, one of the first videos I saw on the channel. I am really enjoying UA-cam recommending me old minute physics videos all of a sudden
builderecks Using which filter? I tried Blur, Gaussian Blur, Motion Blur ... none of which did the right thing. I also tried cubic and sinc interpolation when upscaling and even that didn't do the right thing. That's pretty shocking I have to say. This was all done using Gimp 2.8.10.
samramdebest That's only the interpolation between pixels when scaling the picture. So all but nearest neighbor go through the same colors; just the shape this gradient takes is different. It's got nothing to do with gamma correction. The images in this article explain it much better than my words did: en.wikipedia.org/wiki/Bicubic_interpolation
Penny Lane Don't know if maybe the default settings are different on linux (which is what I use) blur, Gaussian and motion all smoothly blended with no darkness issues in the color test I did.
3:06 you know how many art tutorials there are that tell you to brighten mid tones and blended color areas with arbitrary fake reasons? Sometimes ambient occlusion and subsurface scattering are terms thrown into it, when it's actually just this^ sometimes.
O'kay, YT, I have absolutely no ideas why you are giving me four years old video, but this is actually bloody awesome one! Good job on making this, mate
Can I vote to see a sequel to this video explaining "color space"?? Relating to monitors, tvs, and any digital (or non-digital) final presentations. It would shed more light on the subject... "Color space" can be hard to get your head wrapped around: What are you working/editing in? vs. what is the final output in? and how to compensate appropriately.. The sheer amount of different "color spaces" reminds me of the frustration in the amount of different video codecs there are... which could be another interesting topic/video to explore..?? #danerwinfb
chaquator Colour rhymes with 'duller' ...lol rhymes with log. You're not going to say color. The second o in colour is never pronounced as the o in log.
In Photoshop, when creating a new image, set "Color mode" to "Lab color". That'll set it as the default for new files. When saving to PNG or JPEG, you'll need to go to Image > Mode and set it to RGB.
Checking "Blend RGB Colors Using Gamma" seems to only work for painting (on RGB Color Mode). When I blur the image, I still get the black edges. Switching to Lab Color produces correct results both with painting and blurring whether "Blend RGB Colors Using Gamma" is checked or not. Using Adobe Photoshop CC 2017.
What the heck is Lab Mode, why does everything look better lol (now when you desaturate, the black and white values picture look actually correct). Why do we even use RGB mode?
I think you missed the point of the video. It's not just iOS, it's the vast majority of computers. Every one needs to change, not just the OS you dislike.
Thank you! That was an awesome video to go with my coffee. I learned something and now I want to figure out which editing programs will give me those gorgeous RGB blends!
Years of programming, including reading about gamma, and I never saw mention that both cameras and monitors used logarithmic scales, therefor all our beloved 8-bit image brightness is also on a logarithmic scale. "Gamma correction" is always portrayed as a funky post-processing effect to manipulate brightness, not an intrinsic step the monitor does to reverse what the camera did.
I know this is a old video but can I just say it's AMAZING how easy this is to understand. I have a very shitty range of skills in maths. I do not understand what square roots,timestables are like rocker science to me,ect ect yet despite this I can still understand what your saying. Good job on the way this was worded!
from a programming point of view though, blurring functions are already computationally heavy, and square roots are notoriously slow to process as well. I think we'd find that blurring images the correct way on devices like iOS with high pixel densities might actually produce upsetting lag in the interface. It's the kind of trade off that can be well worth it for the small number of people it might actually upset and teh small number of images it might mess up. IMO
You literally pre-compute the inverse gamma curve and the normal gamma curve...multiply the working texels by the appropriate value of the inverse curve to get back to linear color space, blend, and multiply by the gamma curve to convert back to sRGB encoding....the curve is the same for each color channel too so you don't even have to waste memory pre-calculating for every possible color
well, indeed it's not sqrt, but gamma transform ^1/2.2, or indeed it is sRGB transform that is more complicated. Ok, it can roughly be approximated by sqrt, but please don't say it IS sqrt. It's not more complicated to do the real math.
It doesnt particularly matter, sqrt is just a function that will space big values apart more than it will which is what the video wants to show. Introducing the actual real math there wouldnt serve to do anything, other than alienate the average viewer for no real reason. He puts an asterix for people like you as well
@@dlwatib What is the point of your comment other than to imply that 2.0 is not in between 1.8 and 2.2 (which the video explicitly shows at 2:01)? This is not a rhetorical question.
At 1:27 it's actually not 0 or 1, it's 0 or 255. Every color has one byte assigned so you get only 256 values for each color (0-255). Only value that's stored as a float (decimal) is transparency that's 0.00-1.00
Wouldn't this just create a new problem for any image *not* taken with a digital camera? Are do all current image containers utilize the same squaring algorithms? _Example, an image created solely in Photoshop_
It's not about whether you took the image with a digital camera. It's about the format you store it. Even when you create images in photoshop and then save them as JPG or whatever format Henry is talking about, they will be stored the same way as digital photos.
***** I would assume that formats that are considered less "lossy" and less compressed would avoid this or at least minimize the effects. Such as saving as PNG instead of JPG
***** From a editing standpoint, you should always work with RAW image. The quantity of information it contain REALLY does make a difference. However, even when working with RAW, trying to blur something using RBG does gave you the same dark effect we try to avoid. The LAB color space, as far as I could say, is really the only thing that have a significant impact on this. After all, it doesn't mater how much data you have for an Image if, to begin with, the way the data is altered (editing) is wrong and this is exactly the problem we have here. The problem is not the data, it is the way your program (ex:Photoshop) modify the said data.
All images created for display in the web are created in the sRGB color space, and therefore follow the sRGB gamma curve (roughly a power of 2.2, not technically "squared"). This is so that they don't need modification in order to be displayed by web browsers. JPEG files are generally assumed to be in sRGB, and this is what the average image editor will assume as well. PNG files actually have a gamma and color profile setting so you can store it with any gamma curve you want, but many web browsers and image viewers still horribly suck at proper color management.
D: I had never considered there to be another possibility. Is there a good way to fix this in Photoshop? To be fair to us artists though, having it set this way probably makes it easier for us to transition from physical pigment mediums (paints, coloured pencils, anything of the sort) and better predict the results, since the mix of pigments produces darker, less saturated colours too. It's natural for us to understand the mixing of red and green (or any contrasting, complementary colours) as something that produces dark, desaturated brown. And the method you showed seems to have the problem of generating too much light between the colours, which could prove to be very tricky to deal with for, say, digital illustration. I'd have to test it myself.
+Foxeste You can see it briefly in the video. In Photoshop, click Edit > Colour Adjustments. A window opens up and there should be an option to "blend RGB colours using gamma", with an unticked box and a field where you can enter a number between 1-2,20 (1=most gammafied, 2,2=normal). (I have it in Spanish, so the wording might not be exact.)
This is an interesting problem I had noticed, yet not really pondered much about. Thanks for explaining the reasons for it. Square roots matter more to me now.
if you REALLY want the Blur to use the correct Luminosity value, don't use RGB, but switch to LAB image mode (Image>Mode>Lab in Photoshop). Only there you will find the correct Luminosity applied to the color edges.
In Photoshop I set the image to 32-bits/channel mode, then it does the math right. Too bad many functions aren't implemented or poorly adapted to that mode.
Using Lab is not more "correct", from a physics point of view using linear RGB is correct while Lab is wrong. However, Lab may very well _look better_ since the Lab color space is designed to model human perception.
Just because he mention iOS doesn't mean windows android works different. As you can see he only mention Instagram even those every website works the same way. Morons. He just used something he know people are familiar with
Apple isn't stupid, they picked the inaccurate method for a reason, it's fast. Squaring and then square rooting takes up a lot more processing power, if they would have went with that it would have been laggy. I did some tests and the square root method of finding averages was 30x faster. Edit: someone pointed out to me that using lookup tables (essentially a long list of per-caclulated values) can speed up the squaring method. I tried that and it really speed it up. Using lookup tables the squaring method is now only 2.6x slow, which is a performance hit IOS developers could handle, so yes they are lazy.
I would just add a bright color inbetween the red and green boxes, since i face this problem allot when drawing and am trying to find a solution for when it happens, i just make sure its not too visible for when im done drawing it and zoomed out.
Yes, the alpha channel has the same problem. Do the following experiment: 1) Create a new image in Gimp, filled with red. 2) Create a new layer, half-filled with green. 3) Gaussian-blur the green layer.
I'm beginning to think that framebuffers should be using floating point numbers representing luminances in CIE color space and converted to monitor color space for display by the graphics card. That would make doing the Right Thing so much simpler.
Im a graphic designer with a great interest in pixel art (as you can see from my profile picture). When I first began making pixel art, I was frustrated at how difficult it was to transition a darker color with a different lighter color. I would try to make the gradient logically by making each pixel the same % difference in brightness but the lighter part of the gradient was always almost unnoticable yet there would be a sudden dark line where you could see the difference even from far away. It took me many tries to realize that to make pixel art shading correctly, I would need to make changes in darker areas with extreme precision. Thank you for making this video because I always wondered why I would need to do this.
The computer probably doesn't know if you're dealing with a photograph or regular image. I don't know, but I assume many, if not most image files out there are not photographs.
starship77 did you watch the video? he just finished explaining that the math of most blending operations is wrong because it doesn't take into account the logarithmic scale.
I'm interested in trying out the difference between these settings in photoshop, but even after I made the change shown in this video, photoshop appears to still blend the exact same way. Clicking that box hasn't altered any of the blending methods I've tested so far. Maybe I'm missing something and there's more to it than just clicking that box in the advanced color settings?
Yes., Go to the XML editor, find the filter definition (under ), find the style attribute and remove the 'color-interpolation-filters' property from it (or delete it whole if it's the only property.) Remember that Inkscape can only do what SVG can, and probably less. More info: www.w3.org/TR/SVG/filters.html#FilterPrimitivesOverviewIntro
Great I am sure I watched this video at a certain point in time but now that I am learning game dev and wanting to understand gamma correction for textures I watched it again and it makes more sense now.
+MinutePhysics videos are probably the only UA-cam videos that fuck with my sub, playing a bass line at a frequency it does not like! Grrrrrrrrrrrrrrrrr
That must be annoying, I completely know that feeling. Where something has an odd harmonic that creates a low growling or distorting sound. Wonder what it is. Our town's radio station has a imbalance of 15% on the stereo channels and that already drives me crazy. My right ear is happy, the left one is sad. XDDD Maybe I should email them, I noticed it isn't as bad on stereos as portable devices like phones and MP3 players. Perhaps nobody even knows it does that. Though I do find old analog broadcast hardware quite cool, they have quirks from time to time that's for sure. :D
+theLuigiFan0007 have you tried listening to that station with different radio equipment and/or in different locations? Stereo FM transmission is not trivial. It starts by transmitting the sum L+R, to be compatible with non-stereo receivers, then computes the difference L−R, uses it to amplitude modulate a higher frequency signal, called a Subcarrier, then merges it back with the main signal. What I mean is that there may be some interference in your specific location and/or a fault in your own equipment that gives that imbalance. It may or may not be the station's fault. This is also one of the reasons most stereo equipment (used to?) have a Balance knob, to tweak the stereo balance manually.
etatoby Yeah I know how stereo broadcast works to some extent, isn't the MPX subcarrier between 19kHz to 39kHz? Could be interference, as the roof is a steel roof, which is made of enameled steel plate. But, I don't think so as if I use a USB SDR stick or a car radio there's no imbalance. Both of those auto adjust stereo balance, as far as I know. Could just be older receivers don't like the signal output by the station. I tried it on a somewhat decent stereo a while back and it sounded fine as well. I think the problem is limited to cheap FM radios.
Don't change the settings. Just use LAB color space. That is the *only* correct solution. Setting blend-gamma to 1.0 is wrong wrong wrong wrong wrong. Don't do that. MinutePhysics might know a lot about physics, but he's just sorely mistaken on this point.
***** It's a fundamentally flawed way of editing images. If you start with that as your base, there will always be problems. Either the gamma is too high and you get blurring problems, or the gamma is too low and you get sharpening problems. If you change gamma around continually, you will reduce the image quality (taking a non-standard root involves a lot of numerical errors which are not symmetric on reversal). You will either get stuck in a tangled mess or you will degrade image quality. Colin Richardson Proof? This is basic textbook graphic design material. Linear workflow, LAB colorspace, gamma corrections, etc. It's like asking a chemist for proof why glassware should not be cleaned with a commercial dishwasher.
Interesting, all this time I attributed this to thinking that the perception of color brightness was based on the highest color value (so, which RGB value is exciting our optic cones the most), and so blending red 255,0,0 and green 0,255,0 got a yellow 128,128,0 that appeared dull because its top end was only at 128, even if it had the same total # of photons (or so I thought). The more you know.
I missed the footnote about it actually being a power of 1.8-2.2 rather than just always 2 the first time. Discovering the actual, rather strange at first glance, gamma correction used most often was somewhat unexpected. I think png files use a function that doesn't look like x^2, but rather a small linear part at the start then x^2.2, often approximated as just x^2.1 or something along those lines.
How he got through this video without ever once mentioning that one digital picture (that shall not be named), I'll never know... ;P But seriously, very interesting video! :D
it's actually more impressive how he did not mention the city lights picture from Nasa... scale it without converting to LAB colorspace first, and you get an image that's waaaay different.
***** I guess the correct terminology is transform coding (which is a sub-category of lossy compression techniques): en.wikipedia.org/wiki/Lossy_compression#Transform_coding
Thanks for explaining gamma in a way that doesn't make my brain crust over. I enjoy graphics algorithms and the Gaussian blur issue here is very useful to know!
In display drivers development we end up converting to Linear RGB a lot through a de-gamma process to avoid a lot of the artifacts that occur when blending multiple layers. I wish I knew why a lot of software takes short cuts that look bad.
It's a poor description, I spent a few confused seconds trying to figure out what the hell he was on about before realizing he was talking about gamma. The correct equation is Vc = Vo ^ (1-gamma), and since gamma is around 2.2 on NTSC and most monitors the power value is 0.4545 which is reasonably close to the sqrt function (x^0.5). Gamma does vary considerably though across color systems and devices, so if you're designing art for a handheld console (say) and you want it to look the same on the target then you actually have to design it on a calibrated monitor, transform it by the gamma function of that monitor and then again by the inverse gamma function of the target device.
Because there are infinite data points as you approach zero of its log2() and since computers definitely can't store infinite information, ()^2.2 is the better option (not to mention that log(0) is undefined). Although the decibel scale has all the same problems but is in use.
You know, a fun game for videos like these is to scroll to the comments section and see how many comments there are before some smart-ass tries to prove the central point of the video wrong. For example, this video had a score of five comments.
The thing is, it's arguably bad design to do this gamma correction *during color blending/blurring* in each application via a setting. That may be why even knowledgeable programmers are reluctant to do it, let alone enable it by default. It *should* be done when *reading* the image file, then all the mixing and blending should be done linearly, and then the reverse gamma correction should be applied to the result. Doing gamma correction during color blending would also make the program run slower -- color blending is often done on the GPU these days, and the hardware color blender even in modern GPUs is "fixed-function", i.e. it isn't freely programmable, it's only somewhat parameterizable, and it can only do linear blending (see www.opengl.org/sdk/docs/man/html/glBlendFunc.xhtml). You can do arbitrary non-linear blending in a pixel shader (a small user program that is run by the GPU in parallel for each pixel of the frame to compute its color), but that's still slower than the fixed-function blender, and it may require a large refactoring of your rendering pipeline code, so you may not be able to do that easily. As mentioned, the right way to solve this problem is to apply ^(1/Gamma) to all pixel values of an input image as you're reading it, since common image formats like JPEG have already been gamma-corrected during creation, i.e. their pixel values correspond to (physical light intensity of the pixel)^Gamma. Then you would do all the color mixing and blending linearly, and then apply ^Gamma once to the final result pixel right before sending it to the display hardware -- which internally computes ^(1/Gamma) to get the actual physical brightness of the pixel on the display (this was due to physical properties of the screen on CRTs, and modern LCDs emulate it for backwards compatibility). This is all supported in hardware these days: the sRGB texture format (since OpenGL 2.1) takes care of the gamma correction when reading images -- i.e. it computes texture value = (image pixel value)^(1/Gamma) when reading an image into a texture. Then you can blend multiple texture values linearly, i.e. just add/average them to obtain the result pixel which you write into the framebuffer. After that, the sRGB framebuffer format (GL_FRAMEBUFFER_SRGB, sine OpenGL 3) takes care of computing output pixel value = (framebuffer pixel value)^Gamma, which is sent to the display. I don't know which part of that pipeline common tools like Photoshop or GIMP get wrong (I'm not really an expert on this stuff either), but the solution proposed here ("Blend RGB colors using Gamma" setting in Photoshop) is not optimal.
That's not true. Desktop GPUs have fixed-function blending. Shaders are for pixel generation during rendering, but the blending process that puts those pixels onto the framebuffer is still fixed-function. This is actually programmable on some mobile GPUs which use tile-based rendering, but desktop GPUs do not have this feature. You can simulate arbitrary blending by using more complicated shader programming, but it makes everything a lot more complicated and slower (in some cases impractically so). Olaf is correct. Gamma correction should be considered a *file encoding* detail. In fact it's almost the same as a-law and mu-law audio encoding, and nobody is dumb enough to try to do audio processing with a-law and mu-law data. We just do it with video/graphics because we're used to doing it wrong. GIMP 2.9 (devel) does this properly. Its image operations all use linear 0-1 values. The working image can be internally stored in any format in memory, and in case of gamma-encoding it will be converted to linear light before any operations are performed. For ideal precision across multiple operations, you should change the image mode to 32-bit float linear, work on it, then only export to sRGB the final image. But even without doing that, operations will be correct (though you may lose a bit of precision due to intermediate conversions).
+Christopher King I guess because humans are not machines and everyone perceives colour and brightness a little bit differently than other individuals, and sqrt2 is close enough to the varieting median. And logarithmic is just a general term to describe exponential scaling, sqrt2 can be considered to be fairly logarithmic.
+Christopher King Probably because logarithms are more math-intensive for computers than square roots and the approximation is good enough for our eyes.
This laziness isn't even that hard to undo. You can include a couple of lookup tables in your program containing the correct gamma correction values and get it right very cheaply.
The most common reason I'd see why many computer softwares don't do the 'proper' blurring is that, theoretically, it's much faster just to use the already-present color values instead of squaring then square-rooting every single one of them. That adds a lot of processor overhead, especially on larger images...
Just one question: If you have an RGB value of (255,85,0), is there a third as much green light as red light (i.e., overall brightness is square rooted, whilst proportions are preserved) or is there a ninth as much green light as red light (i.e., each individual brightness is square rooted)? If it's the former, how is the overall brightness calculated?
10se1ucgo Okay. All the colors I have seen, though, are /255, since it is easy to store in a byte (2^8-1). Even if it takes it in decimal, it's very likely to convert it into /255 first and then convert it to binary.
+Shaheer Syed correct me if I'm wrong, but isn't /255 exactly 1 byte, since a byte is 256 bits, and 255 is 256 when you count from 0 instead of 1. Which is how programming languages count.
Images are typically only stored with 8 bits per color channel. At this bit depth you need the gamma "trick" to avoid banding and get smooth gradients. It is all about "cutting corners" to drastically reduce file sizes and processing requirements for images.
***** doing it in image processing software makes sense but when bluring for effect at runtime you have to do it per pixel so that can slow things down for no good reason
Ben A True, but then for real-time applications it might be ok to approximate. Say with a lookup table and some linear interpolation. And for image processing software there really is no excuse to not do this properly.
Penny Lane We do it correctly, search for "linear workflow". Every serious professional that works with images knows what gamma is and why it's important to always work in "linear". All professional software today allows you to work that way. :)
***** blur is pretty expensive effect by itself, which needs to be calculated every frame, to a 60Hz update you only have 16ms before every draw. for HD screens you have million pixels times 3. It is a trade off, you trade visual accuracy to efficiency. For OS animations it is ok to do it that way but for a editing program like photoshop it should be done in the correct way even if it takes more time to compute.
ahh!!! this means a lot to me as a digital artist!! thank you for explaining why my drawings turn out like garbage when i try to blur them
Me too, it really help me as a digital artist :D
I don't understand how it helps. What do you do to fix the problem?
futurestoryteller They could find an application that properly blends colors, such as the aforementioned settings with photoshop :p
In Photoshop, you create a custom RGB setting with a gamma of 1.0. Edit>Convert to Profile. Click on the Profile drop down menu and select Custom RGB. Then type in a gamma of 1.0. You may also want to change the Primaries to Adobe RGB 1998 (mine defaults to HDTV for some reason).
As for any other programs, couldn't tell ya.
If youre trying to blurr little dots or something like that, lower the opacity and lower the size of your smudge tool.
I'm a professional illustrator, and you just taught me something. I'd always assumed it was because image editors were intentionally treating colours like pigments and mixing them subtractively instead of additively, since the result generally fits.
Interestingly, the non-linear storage of brightness in computer graphics did not evolve as a clever deliberate choice; instead, it was merely a legacy from the display systems used back then: Cathode ray tubes. Their brightness happens to be roughly proportional to the square of the control voltage. Designers of TV broadcasting norms were aware of this, and decided to compensate for this effect in the broadcasting side of the system, to keep the receivers as simple as possible. When those same receivers were later adapted as computer displays, the computer engineers never seemed to have paid any attention to this detail. It was only when computers started to be used in the printing industry that this quirk started to get any attention in computer technology.
CLipka2373 Very good
It turns out that while it was a by product of the physics back in the day, It has stuck around because it is actually useful for data compression.
@@TuckerDowns how is it useful for data compression; I've never fully understood that point.
@@brod515 Roughly speaking, taking a square-root means keeping the first half of a numbers most significant bits and dropping the other half, essentially cutting the size in half. This is more complicated in practics, but I how you get the idea.
@@Ruhrpottpatriot This still doesn't quite make sense. information like that would be stored in 4 byte floating point numbers which will still use all the bits to represent a number. I don't think that's what he was referring to as compression... there is a common idea that storing the values non-linearly stores only the useful information for the human eye and I don't quite understand it.
This actually helped me so much with the raytracer I was writing
+chasenallimcam I am a programmer who did not know this, but experienced this before. Now I know why, and will surely forget before i need it again.
+Richard Smith give me your email ill set up a spam program to remind you every 3 hours or so. anything so i dont have to fix your mess later.
My reasons are different, generating color gradients mathematically for display in RGB, the color space of the frame buffer.
hal hahah ok!
Wow, that's dope
As a computer programmer, I think this is less an issue of laziness and more an issue of not realizing the color values were square-rooted in the first place. Thanks for sharing this information.
>An Adobe product not having the default option be the best choice.
As typical as the sun rising in the morning.
haha hating on adobe cause everyone does it how funny and original
@@ForfunckleStudios Hating a company due to their bad consumer practice is clearly wrong
same por Apple
Nice greentext.
@@ForfunckleStudios Hey I've been hating on Adobe ever since Photoshop 2.5 you whippersnapper!
I noticed how as an artist I never use even a tenth of all the very bright white values available to me, and it irritates me a lot when my dark grey gets 1 unit closer to black but it looks much darker...
My mind is blown once again. Thank you, MinutePhysics.
268 likes epik
Gamma correction is like daylight savings time. The actual mathematical operation is super easy, but I can't for the life of me figure out whether to do the one step or its inverse. I keep rewatching this video time and again.
Spring forward, Fall back.
the bass in this video is shaking my house
it ruined my sub
cyclone8200 I see what you did there.
You should check your sub's volume...
INVISIGOTH Bangin' tunes mate. Amirite?
This comment was so random I had to lol.
Minute Physics
Minute Maths
Minute Biology
Minute Technology
4 Minutes of Awesome
Where is minute chemistry?I am a chemist
casaverdero there isnt
casaverdero "In science, there is only physics, all the rest is stamp collecting" -Ernest Rutherford
Ruben Lucescu But we need math for physics....
Regnor Math isn't a science (there's nothing empirical about it)
What is meant by that quote is that at the time physics was a reductionist use of math make testable predictions, most other sciences were still concerned with just phenomenology.
that moment when your drawing program has blurring on an image but it knows what it’s doing and doesn’t make it ugly
Wow! Just checked and Photoshop does mess it up :( But! Blender's node editor, free 3d software, makes it yellow how its supposed it be! :D
*****
Thats ok :)
+Abraham Animations Yup Blender is awesome like that. XD
+Abraham Animations Comparing Blender and Photoshop doesn't make sense tho. They're both made for vastly different reasons.
*****
True, but in the sense of blurring, photoshop doesn't do it quite right :)
Abraham Animations Does it just fine, if you toggle the right mode. The fact that barely anyone noticed this thing prior to this video just speaks volumes how little it matters. And for those whom it does matter (See graphic designers, Texture artists, etc) Photoshop has the option right there for them already, even back in the earlier incarnations of the software, because they know who uses it.
WOW, that was WAY more relevant to me as a photographer than I thought it would be when I clicked on the video
thank you. Now all I see is incorrect bluring.
:P
As a student of computer science I can say this is accurate. We learn to blur images with the wrong approach and then with the good approach. It's about understanding how computer graphics work, the same for bubble sort, we learn the easiest method first. What is wrong is have the wrong method in professional tool as the video says.
So, can you tell me whether the default RGB-values using approach is wrong? Or is this about actual formats?
+Sebb747 like the video says our human vision can't tell the difference between bright colors but dark colors. So, instead of wasting data storing bright colors we can have a better image by storing the root of the original bright value. It's like the mp3 format, instead of saving inaudible sounds we delete those frequencies.
Angel Alvarado Yes, I'm well aware of this. I'm in CS myself.
But if you do image processing, you usually convert your image into an RGB(A) array which you then work with instead of working with the raw data of whatever image the user chose to supply to you.
My Question was whether those RGB values are representing square roots and are being multiplied down the graphics pipeline or whether this is just a problem for people who choose to - for whatever reason - work with the raw image data.
+Sebb747 You can't know unless you have the data from the original source, take for instance a camera, you can set the gamma values on it but once the photo/video is taken all is stored in the basic RGB(A) values. The same when displaying the image, you can change the gamma values in your TV or screen. The thing for us as developers is how to treat those pixels, you can choose the lazy path and use the mean to "blur" the image or be aware that it's not that simple and you need to consider all cases. Color math is an interesting topic as well.
I stopped learning about IP but there are a lot of resources out there.
Angel Alvarado Guess I'll have to write a test case for my image generation stack. Thanks anyway :)
Nope... beauty isn't the default... laziness is... ask any engineer...
performance. computers haven't always been this fast. the image formats come from a time where desktop computers were slower than your average smart phone. he was even making this point in the video.
***** That was in the past... however in software in the present still use the same technique... answer... laziness... trust me I'm first year in IT engineering...
Are you kidding? You think that you can speak for all engineers because you are a first year IT engineering student? Don't make me laugh.
caramida9 Engineers should develop better alternatives for jpeg, png and bmp since they're way outdated.
VirtualCoder Except they already exist and nobody uses them.The same way there has been an alternative to .docx that's one trillion times better and nobody uses it.And the same way averyone should be using .webm instead of .gif but then again nobodo does.
As a (hobby) programmer, I come back to this video occasionally to remind myself about this. Thank you. On this watch I realized I've programmed contrast wrong in a few programs.
More than one million views and no comments??
As a person who does does a lot of graphic art on the computer, I'm amazed I didn't know this before. And why this has not been fixed.
***** Yes. I see them now. I did find it very strange.
***** Thanks for the tip. I will try that.
+Mason Bially In general, the downsides of open software with regards to bugginess, lack of support, and most importantly, lack of economic incentive to fix problems, far outweigh any theoretical advantages.
_"Also, you as the user can always fix the problems with open software."_
Absolutely not. Almost no users are competent to fix problems with any large software package - even if they happen to be professional programmers - that stuff is just a pipe dream in 99% of cases, possibly slightly better if the problem can be fixed by writing a plugin.
The most laughable part of it though, is that if I, as a programmer, ran into a problem with an open source image processing package and knew what the problem was, it would almost certainly be 100 times quicker for me to write my own special purpose program to just do that particular job instead of spending days or weeks attempting to understand the original program to the point where I could modify it safely without breaking other things.
The typical case is that open source software is written by a handful of dedicated enthusiasts, with minimal programming input from users. Blender is a prime example of this, as large critical portions of the program is developed by one guy.(I know this because the lack of development effort prevented me from using the software at one point and it was decidedly not worth my time to write the software myself when I could just buy it from an actual business)
+Julien12150 that happens to me all the time on my phone
The reason it hasn't been fixed is for larger images, the ammount of time a proper blur takes is far more meaningful then smaller images. a 1024x1024 image has just over 1 million pixels.
The first method uses an addition and division operation per pixel (The colors are already square rooted), So for the picture using the first method, just over 1 million additions and divisions
The second method uses 1 addition, 1 division, 2 multiplications (squares), and 1 square root (The most expensive basic math function a computer can do(Not counting trig functions)). So this multiplied by 1 million, and it would take 1 million additions, divisions, and square roots, and 2 million multiplications.
If you are going for a faster program with less wait time between blurring operations (paint, photoshop) or less intense software on older hardware in general, you go with the 2 million operations rather then the 5 million operations
He just basically called Apple lazy :D
+minecraftace123 Ya apples are lazy they just hang on trees
+Ethan Chou How very true. . .
The level of polish we've come to expect from Apple products
You're right, like i'm totaly sure the apple engineers just accidently put 2gb or DDR2 RAM in a computer that has 2 4GHz quad core processors. It totaly wasn't just to scam idiots out of their money or anything.
LE/A Tyrone Indeed, indeed!
Damb I remember when this video was new, one of the first videos I saw on the channel. I am really enjoying UA-cam recommending me old minute physics videos all of a sudden
Is there a setting for this in GIMP?
I want to know the same thing, I think GIMP does this because i found the settings, cubic and linear. (with standard cubic)
On my copy of gimp it did it right by default.
builderecks Using which filter? I tried Blur, Gaussian Blur, Motion Blur ... none of which did the right thing. I also tried cubic and sinc interpolation when upscaling and even that didn't do the right thing. That's pretty shocking I have to say. This was all done using Gimp 2.8.10.
samramdebest That's only the interpolation between pixels when scaling the picture. So all but nearest neighbor go through the same colors; just the shape this gradient takes is different. It's got nothing to do with gamma correction. The images in this article explain it much better than my words did: en.wikipedia.org/wiki/Bicubic_interpolation
Penny Lane Don't know if maybe the default settings are different on linux (which is what I use) blur, Gaussian and motion all smoothly blended with no darkness issues in the color test I did.
3:06 you know how many art tutorials there are that tell you to brighten mid tones and blended color areas with arbitrary fake reasons? Sometimes ambient occlusion and subsurface scattering are terms thrown into it, when it's actually just this^ sometimes.
O'kay, YT, I have absolutely no ideas why you are giving me four years old video, but this is actually bloody awesome one! Good job on making this, mate
Can I vote to see a sequel to this video explaining "color space"??
Relating to monitors, tvs, and any digital (or non-digital) final presentations. It would shed more light on the subject... "Color space" can be hard to get your head wrapped around: What are you working/editing in? vs. what is the final output in? and how to compensate appropriately.. The sheer amount of different "color spaces" reminds me of the frustration in the amount of different video codecs there are... which could be another interesting topic/video to explore..??
#danerwinfb
I can't believe you made a video for this.... hahahahahah AWESOME!
2:09 If the rooting part gave us the results we wanted then why do we square the numbers anyway?
Americans blend away the u in color.
*colour
lol
i bet brit bongs say "loul" instead, too
chaquator
Colour rhymes with 'duller' ...lol rhymes with log. You're not going to say color. The second o in colour is never pronounced as the o in log.
The u in color is like the brown in between two colors, ugly and not needed
Innar Koït Chtofenbeurg AmE - color BE - colour
I FINALLY understand the purpose of lab color mode in Photoshop! thank you
Thanks!
👍
In Photoshop, when creating a new image, set "Color mode" to "Lab color". That'll set it as the default for new files.
When saving to PNG or JPEG, you'll need to go to Image > Mode and set it to RGB.
Checking "Blend RGB Colors Using Gamma" seems to only work for painting (on RGB Color Mode). When I blur the image, I still get the black edges.
Switching to Lab Color produces correct results both with painting and blurring whether "Blend RGB Colors Using Gamma" is checked or not.
Using Adobe Photoshop CC 2017.
What the heck is Lab Mode, why does everything look better lol (now when you desaturate, the black and white values picture look actually correct). Why do we even use RGB mode?
@@Fynmorphover cielab?
Ill add this to my giant list to why ios sucks
iOS really doesn't suck....iOS just serves a different purpose from Linux and Windows.
***** Well iOS is basically a very expensive version of unix so...
Hunter Grimx This isn't limited to just iOS...
I think you missed the point of the video. It's not just iOS, it's the vast majority of computers. Every one needs to change, not just the OS you dislike.
***** It's overpriced as fuck.
Thank you! That was an awesome video to go with my coffee. I learned something and now I want to figure out which editing programs will give me those gorgeous RGB blends!
Beauty is the default! Look how elegantly an entire image was stored using as few bright gradations as the human eye can even notice!
Years of programming, including reading about gamma, and I never saw mention that both cameras and monitors used logarithmic scales, therefor all our beloved 8-bit image brightness is also on a logarithmic scale. "Gamma correction" is always portrayed as a funky post-processing effect to manipulate brightness, not an intrinsic step the monitor does to reverse what the camera did.
I know this is a old video but can I just say it's AMAZING how easy this is to understand. I have a very shitty range of skills in maths. I do not understand what square roots,timestables are like rocker science to me,ect ect yet despite this I can still understand what your saying. Good job on the way this was worded!
this explains alot
i thank you for letting us know this
Anybody know where the setting is for Gimp?
Dr. Certifiable somewhere underneath the leather suite.
sadly incorrect, just tested ^^
Blargles Malargles
"Leather suite"...? Idk what that is but Kaveh is saying that's not right. Can you clarify?
Dr. Certifiable I think it has to do with BDSM. www.urbandictionary.com/define.php?term=gimp
Dr. Certifiable I think you have to search under Tools -> GEGL libraries -> gausian-blur, but I'm really not sure..
2:53 Fun little thing to try to prove yourself (if you like math).
(sqrt(x)+sqrt(y))/2
Holy shit. I consider myself a software developer with good understanding of image processing, but this is news for me.
Beauty should be the default. That's true in many circumstances.
Too bad most things and most people aren't beautiful.
What a beautiful video.. and a beautiful observation. Well done minutephysics
from a programming point of view though, blurring functions are already computationally heavy, and square roots are notoriously slow to process as well. I think we'd find that blurring images the correct way on devices like iOS with high pixel densities might actually produce upsetting lag in the interface. It's the kind of trade off that can be well worth it for the small number of people it might actually upset and teh small number of images it might mess up. IMO
That's what look up tables are for. :)
You have no fucking clue what a lookup table is.
No, I was not referring to the *blur*, but the gamma correction!
What Joe meant was a rainbow table. You only need a couple of megabytes to map one for every single color
You literally pre-compute the inverse gamma curve and the normal gamma curve...multiply the working texels by the appropriate value of the inverse curve to get back to linear color space, blend, and multiply by the gamma curve to convert back to sRGB encoding....the curve is the same for each color channel too so you don't even have to waste memory pre-calculating for every possible color
well, indeed it's not sqrt, but gamma transform ^1/2.2, or indeed it is sRGB transform that is more complicated. Ok, it can roughly be approximated by sqrt, but please don't say it IS sqrt. It's not more complicated to do the real math.
It doesnt particularly matter, sqrt is just a function that will space big values apart more than it will which is what the video wants to show. Introducing the actual real math there wouldnt serve to do anything, other than alienate the average viewer for no real reason. He puts an asterix for people like you as well
What's the point of the video if not to be accurate?
The video literally says this already at 2:01.
@@dlwatib What is the point of your comment other than to imply that 2.0 is not in between 1.8 and 2.2 (which the video explicitly shows at 2:01)? This is not a rhetorical question.
My sister works for Valspar Paint and creates her CH (Color Harmony) thoroughly through it. I love her!
This is the first time in a long time I could follow along holy shit
is anyone else getting super bass in their headphones?
no i am not getting fish in my headphones. if you are, please see a doctor
lol that actually made me laugh
So you don't laugh at a fish very often.
I'm not really a fan of Nicki Minaj
Someone isn't using neutral headphones I see
At 1:27 it's actually not 0 or 1, it's 0 or 255. Every color has one byte assigned so you get only 256 values for each color (0-255). Only value that's stored as a float (decimal) is transparency that's 0.00-1.00
That's why professional photographers always use "raw" image format. It preserves colour and brightness information correctly.
So I accidentally turned on my translator and it messes up really often so showed a different comment and then this comment
@@banana_man_101 ok
YES BEAUTY SHOULD BE THE DEFAULT! YES YES YES!
1:52 Only one are the same others you marked as "basically the same" aren't the same.
It's super easy for me to blend colors. All I have to do is take off my glasses! (BTW, I'm nearsighted.)
myopia gang
♥️"Shouldn't beauty be the default?"♥️
Simply amazing. Thank you so much for this marvelous work.
Wouldn't this just create a new problem for any image *not* taken with a digital camera?
Are do all current image containers utilize the same squaring algorithms?
_Example, an image created solely in Photoshop_
It's not about whether you took the image with a digital camera. It's about the format you store it. Even when you create images in photoshop and then save them as JPG or whatever format Henry is talking about, they will be stored the same way as digital photos.
***** I would assume that formats that are considered less "lossy" and less compressed would avoid this or at least minimize the effects. Such as saving as PNG instead of JPG
***** From a editing standpoint, you should always work with RAW image. The quantity of information it contain REALLY does make a difference. However, even when working with RAW, trying to blur something using RBG does gave you the same dark effect we try to avoid. The LAB color space, as far as I could say, is really the only thing that have a significant impact on this. After all, it doesn't mater how much data you have for an Image if, to begin with, the way the data is altered (editing) is wrong and this is exactly the problem we have here. The problem is not the data, it is the way your program (ex:Photoshop) modify the said data.
All images created for display in the web are created in the sRGB color space, and therefore follow the sRGB gamma curve (roughly a power of 2.2, not technically "squared"). This is so that they don't need modification in order to be displayed by web browsers. JPEG files are generally assumed to be in sRGB, and this is what the average image editor will assume as well.
PNG files actually have a gamma and color profile setting so you can store it with any gamma curve you want, but many web browsers and image viewers still horribly suck at proper color management.
D: I had never considered there to be another possibility. Is there a good way to fix this in Photoshop?
To be fair to us artists though, having it set this way probably makes it easier for us to transition from physical pigment mediums (paints, coloured pencils, anything of the sort) and better predict the results, since the mix of pigments produces darker, less saturated colours too. It's natural for us to understand the mixing of red and green (or any contrasting, complementary colours) as something that produces dark, desaturated brown. And the method you showed seems to have the problem of generating too much light between the colours, which could prove to be very tricky to deal with for, say, digital illustration. I'd have to test it myself.
Oh, sweet! I found the setting, and the gamma adjustment allows you to avoid the problem of too much brightness going on keeping the setting at ~1.5.
Wait, where's the setting? D:
+Foxeste You can see it briefly in the video. In Photoshop, click Edit > Colour Adjustments. A window opens up and there should be an option to "blend RGB colours using gamma", with an unticked box and a field where you can enter a number between 1-2,20 (1=most gammafied, 2,2=normal).
(I have it in Spanish, so the wording might not be exact.)
This is an interesting problem I had noticed, yet not really pondered much about. Thanks for explaining the reasons for it. Square roots matter more to me now.
if you REALLY want the Blur to use the correct Luminosity value, don't use RGB, but switch to LAB image mode (Image>Mode>Lab in Photoshop).
Only there you will find the correct Luminosity applied to the color edges.
In Photoshop I set the image to 32-bits/channel mode, then it does the math right. Too bad many functions aren't implemented or poorly adapted to that mode.
THANK YOU ALOT. I really wanted correct this because the blur effects I used to ajust ilummination in 3D render was getting a weird darkess. =D =D =D
Using Lab is not more "correct", from a physics point of view using linear RGB is correct while Lab is wrong. However, Lab may very well _look better_ since the Lab color space is designed to model human perception.
And from what I've seen, you can use it to adjust skin tones. (Flesh Man Group intensifies)
I'm gonna blatantly send this video to all iOS and software developers.
Boogster Su I like that that implies that IOS developers aren't software developers.
lol like they're gonna listen to what people actually want
Just because he mention iOS doesn't mean windows android works different. As you can see he only mention Instagram even those every website works the same way.
Morons. He just used something he know people are familiar with
Yes, because they are all incompetent people who know nothing about this problem...
Apple isn't stupid, they picked the inaccurate method for a reason, it's fast. Squaring and then square rooting takes up a lot more processing power, if they would have went with that it would have been laggy. I did some tests and the square root method of finding averages was 30x faster.
Edit: someone pointed out to me that using lookup tables (essentially a long list of per-caclulated values) can speed up the squaring method. I tried that and it really speed it up. Using lookup tables the squaring method is now only 2.6x slow, which is a performance hit IOS developers could handle, so yes they are lazy.
I would just add a bright color inbetween the red and green boxes, since i face this problem allot when drawing and am trying to find a solution for when it happens, i just make sure its not too visible for when im done drawing it and zoomed out.
I had no idea! Is the same true for the alpha channel in semi-transparent "colors"?
Nice!
Yup!
Yes, the alpha channel has the same problem. Do the following experiment:
1) Create a new image in Gimp, filled with red.
2) Create a new layer, half-filled with green.
3) Gaussian-blur the green layer.
I'm beginning to think that framebuffers should be using floating point numbers representing luminances in CIE color space and converted to monitor color space for display by the graphics card. That would make doing the Right Thing so much simpler.
Now i know why Play Station has that weird (but nostalgic) looking fade out.
Trueee
Im a graphic designer with a great interest in pixel art (as you can see from my profile picture). When I first began making pixel art, I was frustrated at how difficult it was to transition a darker color with a different lighter color. I would try to make the gradient logically by making each pixel the same % difference in brightness but the lighter part of the gradient was always almost unnoticable yet there would be a sudden dark line where you could see the difference even from far away. It took me many tries to realize that to make pixel art shading correctly, I would need to make changes in darker areas with extreme precision.
Thank you for making this video because I always wondered why I would need to do this.
The computer probably doesn't know if you're dealing with a photograph or regular image. I don't know, but I assume many, if not most image files out there are not photographs.
***** Doesn't matter.
it doesn't matter. even if you paint on an image directly, you're still using pre-square-rooted color values.
i mean in multiple demonstrations shown in the video, the blurred example images were not photographs.
Roger Levy It's the monitor that is displaying it "wrong". The math is correct.
starship77 did you watch the video? he just finished explaining that the math of most blending operations is wrong because it doesn't take into account the logarithmic scale.
I'm interested in trying out the difference between these settings in photoshop, but even after I made the change shown in this video, photoshop appears to still blend the exact same way. Clicking that box hasn't altered any of the blending methods I've tested so far. Maybe I'm missing something and there's more to it than just clicking that box in the advanced color settings?
Same here :-/
his method will fix it for blending one layer on top of another one.
To fix blurring one layer into itself, you have to use LAB mode rather than RGB.
Not sure if you ever figured it out but this is what I do to fix it:
imgur.com/a/Ovc9bsJ
I didn't know how badly I wanted to know the reason behind ugly blurring.....Thank you!
Hmmm, now can you fix this in Inkscape?
Yes., Go to the XML editor, find the filter definition (under ), find the style attribute and remove the 'color-interpolation-filters' property from it (or delete it whole if it's the only property.)
Remember that Inkscape can only do what SVG can, and probably less. More info:
www.w3.org/TR/SVG/filters.html#FilterPrimitivesOverviewIntro
Do you know the settings for this on gimp ? And btw, great video !
Great I am sure I watched this video at a certain point in time but now that I am learning game dev and wanting to understand gamma correction for textures I watched it again and it makes more sense now.
+MinutePhysics videos are probably the only UA-cam videos that fuck with my sub, playing a bass line at a frequency it does not like! Grrrrrrrrrrrrrrrrr
That must be annoying, I completely know that feeling. Where something has an odd harmonic that creates a low growling or distorting sound. Wonder what it is. Our town's radio station has a imbalance of 15% on the stereo channels and that already drives me crazy. My right ear is happy, the left one is sad. XDDD
Maybe I should email them, I noticed it isn't as bad on stereos as portable devices like phones and MP3 players. Perhaps nobody even knows it does that. Though I do find old analog broadcast hardware quite cool, they have quirks from time to time that's for sure. :D
+theLuigiFan0007 have you tried listening to that station with different radio equipment and/or in different locations? Stereo FM transmission is not trivial. It starts by transmitting the sum L+R, to be compatible with non-stereo receivers, then computes the difference L−R, uses it to amplitude modulate a higher frequency signal, called a Subcarrier, then merges it back with the main signal.
What I mean is that there may be some interference in your specific location and/or a fault in your own equipment that gives that imbalance. It may or may not be the station's fault. This is also one of the reasons most stereo equipment (used to?) have a Balance knob, to tweak the stereo balance manually.
etatoby
Yeah I know how stereo broadcast works to some extent, isn't the MPX subcarrier between 19kHz to 39kHz?
Could be interference, as the roof is a steel roof, which is made of enameled steel plate. But, I don't think so as if I use a USB SDR stick or a car radio there's no imbalance. Both of those auto adjust stereo balance, as far as I know.
Could just be older receivers don't like the signal output by the station. I tried it on a somewhat decent stereo a while back and it sounded fine as well. I think the problem is limited to cheap FM radios.
Maybe you need to use advanced settings ;)
Elf pain how to change the settings for photoshop
Don't change the settings. Just use LAB color space. That is the *only* correct solution. Setting blend-gamma to 1.0 is wrong wrong wrong wrong wrong. Don't do that. MinutePhysics might know a lot about physics, but he's just sorely mistaken on this point.
***** link to proof please?
***** It isn't. It's wrong wrong wrong wrong wrong.
***** It's a fundamentally flawed way of editing images. If you start with that as your base, there will always be problems. Either the gamma is too high and you get blurring problems, or the gamma is too low and you get sharpening problems. If you change gamma around continually, you will reduce the image quality (taking a non-standard root involves a lot of numerical errors which are not symmetric on reversal). You will either get stuck in a tangled mess or you will degrade image quality.
Colin Richardson Proof? This is basic textbook graphic design material. Linear workflow, LAB colorspace, gamma corrections, etc. It's like asking a chemist for proof why glassware should not be cleaned with a commercial dishwasher.
***** No, but you can ask a chemist to see what text book they are quoting
Interesting, all this time I attributed this to thinking that the perception of color brightness was based on the highest color value (so, which RGB value is exciting our optic cones the most), and so blending red 255,0,0 and green 0,255,0 got a yellow 128,128,0 that appeared dull because its top end was only at 128, even if it had the same total # of photons (or so I thought). The more you know.
Well this answers why gradients with transparencies are so ugly in Illustrator.
Audible, the leading provider of UA-cam sponsorship!
I missed the footnote about it actually being a power of 1.8-2.2 rather than just always 2 the first time. Discovering the actual, rather strange at first glance, gamma correction used most often was somewhat unexpected.
I think png files use a function that doesn't look like x^2, but rather a small linear part at the start then x^2.2, often approximated as just x^2.1 or something along those lines.
2015: nope
2016: still no
2017: nah m8
2018:no!
2019: *RECCOMEND THIS NOW NOW NOW!!!!!!*
i love how youtube does that.
@@professormutant3252 Yes. :]
you worthless cretin this was still viewed before 2019
How he got through this video without ever once mentioning that one digital picture (that shall not be named), I'll never know... ;P
But seriously, very interesting video! :D
what image?
Raicuparta The image of a particular item of clothing :P
josh11735
pfft, that fad lasted like 2 hours
cyberconsumer That's why it was a joke
it's actually more impressive how he did not mention the city lights picture from Nasa... scale it without converting to LAB colorspace first, and you get an image that's waaaay different.
brilliant video! understandable and yet in-depth. Finally, I know what the gamma in colour setting is!
I totally agree with that last statement
What an insightful comment!
+Roy Vivat comments don't need to provide insight
What an insightful comment!
Did you even understand that last statement?
So this blurring issue is really an artefact of file compression?
Not really compression, just storing it in a way that has the most value to the human eye.
Craig McIlwrath Thanks kind kerbal!
Craig McIlwrath It *is* a compression! Just not in the pixel data, but in value range of the pixels. (You can do both at the same time, of course.)
Dexter Netwon artifact of converting a log scale in a linear scale.
***** I guess the correct terminology is transform coding (which is a sub-category of lossy compression techniques): en.wikipedia.org/wiki/Lossy_compression#Transform_coding
Thanks for explaining gamma in a way that doesn't make my brain crust over. I enjoy graphics algorithms and the Gaussian blur issue here is very useful to know!
2:34
You missed the part where you draw red! xD
yeah but.....will it blend?
oh, yes. yes it does.
In display drivers development we end up converting to Linear RGB a lot through a de-gamma process to avoid a lot of the artifacts that occur when blending multiple layers. I wish I knew why a lot of software takes short cuts that look bad.
sqrt()? Why not log2()?
It's a poor description, I spent a few confused seconds trying to figure out what the hell he was on about before realizing he was talking about gamma. The correct equation is Vc = Vo ^ (1-gamma), and since gamma is around 2.2 on NTSC and most monitors the power value is 0.4545 which is reasonably close to the sqrt function (x^0.5). Gamma does vary considerably though across color systems and devices, so if you're designing art for a handheld console (say) and you want it to look the same on the target then you actually have to design it on a calibrated monitor, transform it by the gamma function of that monitor and then again by the inverse gamma function of the target device.
Because there are infinite data points as you approach zero of its log2() and since computers definitely can't store infinite information, ()^2.2 is the better option (not to mention that log(0) is undefined). Although the decibel scale has all the same problems but is in use.
You know, a fun game for videos like these is to scroll to the comments section and see how many comments there are before some smart-ass tries to prove the central point of the video wrong. For example, this video had a score of five comments.
For me I got 9 comments in.
8 comments in. Wow, you were pretty much right.
Way to go Henry! Great analysis.
The thing is, it's arguably bad design to do this gamma correction *during color blending/blurring* in each application via a setting. That may be why even knowledgeable programmers are reluctant to do it, let alone enable it by default. It *should* be done when *reading* the image file, then all the mixing and blending should be done linearly, and then the reverse gamma correction should be applied to the result. Doing gamma correction during color blending would also make the program run slower -- color blending is often done on the GPU these days, and the hardware color blender even in modern GPUs is "fixed-function", i.e. it isn't freely programmable, it's only somewhat parameterizable, and it can only do linear blending (see www.opengl.org/sdk/docs/man/html/glBlendFunc.xhtml). You can do arbitrary non-linear blending in a pixel shader (a small user program that is run by the GPU in parallel for each pixel of the frame to compute its color), but that's still slower than the fixed-function blender, and it may require a large refactoring of your rendering pipeline code, so you may not be able to do that easily.
As mentioned, the right way to solve this problem is to apply ^(1/Gamma) to all pixel values of an input image as you're reading it, since common image formats like JPEG have already been gamma-corrected during creation, i.e. their pixel values correspond to (physical light intensity of the pixel)^Gamma. Then you would do all the color mixing and blending linearly, and then apply ^Gamma once to the final result pixel right before sending it to the display hardware -- which internally computes ^(1/Gamma) to get the actual physical brightness of the pixel on the display (this was due to physical properties of the screen on CRTs, and modern LCDs emulate it for backwards compatibility).
This is all supported in hardware these days: the sRGB texture format (since OpenGL 2.1) takes care of the gamma correction when reading images -- i.e. it computes texture value = (image pixel value)^(1/Gamma) when reading an image into a texture. Then you can blend multiple texture values linearly, i.e. just add/average them to obtain the result pixel which you write into the framebuffer. After that, the sRGB framebuffer format (GL_FRAMEBUFFER_SRGB, sine OpenGL 3) takes care of computing output pixel value = (framebuffer pixel value)^Gamma, which is sent to the display. I don't know which part of that pipeline common tools like Photoshop or GIMP get wrong (I'm not really an expert on this stuff either), but the solution proposed here ("Blend RGB colors using Gamma" setting in Photoshop) is not optimal.
That's not true. Desktop GPUs have fixed-function blending. Shaders are for pixel generation during rendering, but the blending process that puts those pixels onto the framebuffer is still fixed-function. This is actually programmable on some mobile GPUs which use tile-based rendering, but desktop GPUs do not have this feature. You can simulate arbitrary blending by using more complicated shader programming, but it makes everything a lot more complicated and slower (in some cases impractically so).
Olaf is correct. Gamma correction should be considered a *file encoding* detail. In fact it's almost the same as a-law and mu-law audio encoding, and nobody is dumb enough to try to do audio processing with a-law and mu-law data. We just do it with video/graphics because we're used to doing it wrong.
GIMP 2.9 (devel) does this properly. Its image operations all use linear 0-1 values. The working image can be internally stored in any format in memory, and in case of gamma-encoding it will be converted to linear light before any operations are performed. For ideal precision across multiple operations, you should change the image mode to 32-bit float linear, work on it, then only export to sRGB the final image. But even without doing that, operations will be correct (though you may lose a bit of precision due to intermediate conversions).
The solution, don't use crappy tools.
Jeffery Liggett or learn how to use non crappy tools
Jason Crafts
That's the correct way to define it.
***** imageMagick 8D
***** MS Paint.
Jeffery Liggett Don't use Adobe software......LOL
Working in Photoshop for 6 years now and learned something new today... thanks!!!
if human vision is logarithmic, why do they use the square root?
All logs are proportional anyway, so why not
+Christopher King Maybe square root is easier to do and in the range 0-1 gets close enough.
+jerielmari Log Base 2(x) is NOT = to sqrt(x).
+Christopher King I guess because humans are not machines and everyone perceives colour and brightness a little bit differently than other individuals, and sqrt2 is close enough to the varieting median. And logarithmic is just a general term to describe exponential scaling, sqrt2 can be considered to be fairly logarithmic.
+Christopher King Probably because logarithms are more math-intensive for computers than square roots and the approximation is good enough for our eyes.
This laziness isn't even that hard to undo. You can include a couple of lookup tables in your program containing the correct gamma correction values and get it right very cheaply.
The most common reason I'd see why many computer softwares don't do the 'proper' blurring is that, theoretically, it's much faster just to use the already-present color values instead of squaring then square-rooting every single one of them. That adds a lot of processor overhead, especially on larger images...
Just one question: If you have an RGB value of (255,85,0), is there a third as much green light as red light (i.e., overall brightness is square rooted, whilst proportions are preserved) or is there a ninth as much green light as red light (i.e., each individual brightness is square rooted)? If it's the former, how is the overall brightness calculated?
Who else saw the eclipse in the UK today?
Who gives a shit....
william singleton Too cloudy D':
We saw it here in the Netherlands too.
If I understand correctly it was most of Europe and North Africa.
I didn't even see a sun today, let alone an eclipse. Too damn cloudy. (Netherlands here)
sogghartha same
It's /255, not /1 (for the most common 8 bit color).
For example, white is RGB(255, 255, 255) and not RGB(1, 1, 1);
+Shaheer Syed Both are fine and commonly used. 1 is better for doing maths with.
Not necessarily. For example, the OpenGL function glColor3f takes 3 numbers from the range 0.0 to 1.0, not 0 to 255
10se1ucgo Okay. All the colors I have seen, though, are /255, since it is easy to store in a byte (2^8-1). Even if it takes it in decimal, it's very likely to convert it into /255 first and then convert it to binary.
+Shaheer Syed correct me if I'm wrong, but isn't /255 exactly 1 byte, since a byte is 256 bits, and 255 is 256 when you count from 0 instead of 1. Which is how programming languages count.
Michelle Gorbunova Thats exactly what I said, 8 bits are 1 byte,
Minutephysics, giving me the answers to questions I didn't even know to ask.
At the risk of sounding cliché, first.
Why would this be turned off by default? It's worse than kerning being turned of in MS Word by default.
Images are typically only stored with 8 bits per color channel. At this bit depth you need the gamma "trick" to avoid banding and get smooth gradients.
It is all about "cutting corners" to drastically reduce file sizes and processing requirements for images.
"typical Ms thing" said on the video where literally most software ranging from adobe photoshop to apple's IOS are shown to do the same thing
You've got a ground loop somewhere in your audio chain - there's a 60Hz tone whenever the mic is active.
square rooting is a very expensive process for a computer and not worth doing in most cases
***** doing it in image processing software makes sense but when bluring for effect at runtime you have to do it per pixel so that can slow things down for no good reason
Ben A True, but then for real-time applications it might be ok to approximate. Say with a lookup table and some linear interpolation.
And for image processing software there really is no excuse to not do this properly.
Penny Lane We do it correctly, search for "linear workflow". Every serious professional that works with images knows what gamma is and why it's important to always work in "linear". All professional software today allows you to work that way. :)
Jack Hudler You're too late ;)
***** blur is pretty expensive effect by itself, which needs to be calculated every frame, to a 60Hz update you only have 16ms before every draw. for HD screens you have million pixels times 3. It is a trade off, you trade visual accuracy to efficiency. For OS animations it is ok to do it that way but for a editing program like photoshop it should be done in the correct way even if it takes more time to compute.