Severe contrast enhancements: on 45% rule — and cats

Started Jun 28, 2013 | Discussions
VicC
Senior MemberPosts: 1,257Gear list
Like?
Antoich
In reply to Antioch, Jul 2, 2013

Why don't you see my post? It is there? Vic

Reply   Reply with quote   Complain
Toermalijn
Forum ProPosts: 12,638Gear list
Like?
Re:Could somebody draw in the cat ! I do not see it !
In reply to Antioch, Jul 2, 2013

Antioch wrote:

Regards

Both ralph and vicc pointed out the cat...

Reply   Reply with quote   Complain
Antioch
Senior MemberPosts: 1,777
Like?
Re:Could somebody draw in the cat ! I do not see it !
In reply to Toermalijn, Jul 2, 2013

Toermalijn wrote:

Antioch wrote:

Regards

Both ralph and vicc pointed out the cat...

Yep, I saw both of the post pointing to it, but I still dont see it, as others have pointed out too. That is why I have asked for it to be painted out. At this point, I have to say, it is an imaginary cat ... lol.

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re: Changing contrast without distorting hue and saturation
In reply to Tom Axford, Jul 9, 2013

Tom Axford wrote:

I sympathise with your frustration.

I do not recollect mentioning any frustration…

I spent a couple of years puzzling over how to apply a tone curve to an image without distorting the colours (both hue and saturation). After doing quite a lot of reading about colour models and experimenting, initially in Photoshop, later in GIMP (I have given up using Photoshop), I found a method of doing it, which is described here .

What you describe in this reference is, essentially, “the 100% rule”: you keep the saturation of the initial image intact. This is the “variant 2” of my post (with the variant 3 being the “the 0% rule”).

The whole topic of my post was that your method gives unnatural colors; hence “the 45% rule”. For my eyes, 45 gives visually better results than “the 40% rule” or “the 50% rule” — which are both significantly better than “the 100% rule” or “the 0% rule”.

And, what you mentioned is trivial to achieve in GIMP (although until I’ve read the source code of “Mode”s, I have been using yet more convoluted way that yours! It is a pity that what GIMP does is completely undocumented). Just

  • Duplicate the layer;
  • Change the Mode of the top copy to “Level”;
  • Apply curve to the top layer.

My method is mathematically correct,

¿¿¿ What mathematical correctness has to do with this ??? Setting all R,G,B to 0 is “mathematically correct”. What is, IMO, the topic of discussion is the human perception of colors after severe contrast manipulations.

assuming I have correctly guessed what the various tools in GIMP do (unfortunately, I have not found any mathematical definition of the tools in GIMP, or in Photoshop for that matter, but the basic ones seem fairly obvious).

GIMP has an advantage that the source code is there, so one can google for this. But the value hierarchy of GIMP developers was always a mystery to me — why spend a lot of time implementing nice features, then skip documenting what they do. So now users must work like script kiddies: experiment/report/do-as-I-do.

However, the method is a little tedious to apply (although you soon get used to doing it). It would be much preferable if GIMP implemented it directly in the tools that operate on the tone curve (Brightness-Contrast, Levels, Curves, etc.).

My thought exactly. Only if my understanding is correct, one would need an additional slider so that “45” of “the 45% rule” is not hardwired. (In my experiments, 45 was consistently the best; but such things are very often subjective and depend on a person.)

If someone experimented with this: did you find other numbers more suitable in some situations?

In fact, it would be a minor change to the program code to implement it. I'm sure it will happen sometime in the future, let's hope sooner rather than later.

I would not be so hopeful — I have some experience communicating with GIMP developers. (One gets an impression that they think they have much more clue than what it looks from outside.)

By the way, I'm not sure that this helps in finding the cat!

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re: Severe contrast enhancements: on 45% rule — and cats
In reply to gwlaw99, Jul 9, 2013

gwlaw99 wrote:

Would converting to LAB mode and only applying contrast enhancements to the L channel work?

As far as colors are concerned, there is no practical difference between this and my “variant 2” (or Tom’s method). Observe: what you propose is to keep the chromaticity intact. The only difference is that here you interpret “brightness” (of the brightness/chromaticity) pair as “luminance”, not as “level” (which is max(R,G,B)).

Comparing application of a curve to the level, and application to the luminocity, I may imagine that the results may be visually distinctive (but I expect the distinction to be very minor). On the other hand, if you keep the chromaticity intact, this distinction will not affect the chromaticity part of the image.

(And if it affects the “perceived chromaticity”, then only because the оf the mismatch between the “calculated chromaticity” and “perceived chromaticity”; and, AFAIK, this mismatch is not significant. So the “minor distinction” is multiplied by an “insignificant mismatch”. This is why I wrote “no practical difference” above.)

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re: LAB mode does not preserve the colours
In reply to Deleted1929, Jul 9, 2013

sjgcit wrote:

The L in LAB is not a linear mix of RGB. It's considerably more complex.

True, but irrelevant in the context of my question (keeping the colors while changing the contrast a lot).  Since L is a function of Y, applying a certain curve to L is fully equivalent to applying (another) curve to Y.

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re:Could somebody draw in the cat ! I do not see it !
In reply to Antioch, Jul 10, 2013

Antioch wrote:

Toermalijn wrote:

Antioch wrote:

Regards

Both ralph and vicc pointed out the cat...

Yep, I saw both of the post pointing to it, but I still dont see it, as others have pointed out too. That is why I have asked for it to be painted out. At this point, I have to say, it is an imaginary cat ... lol.

Your feelings are very familiar to me (see the OP).  However, after doing enhancements, and finally finding the beast, now I can’t NOT see it.  Here is the central (yes!) crop of the enhanced image from the OP (cat is walking our way, or maybe ↘ way):

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re: LAB mode does not preserve the colours
In reply to knickerhawk, Jul 10, 2013

It seems like kind of an odd argument that adjusting L in LAB "does not preserve the colors". By definition it does.

You speak as a humanities guy lured by the magic of formulae. As opposed to that, in exact sciences we treat formulae as just “an acceptable approximation to reality”; and the reality in the frame of this discussion is the human perception of colors.  AFAIU, we know close to zilch about this topic. (BTW, I started this discussion as a way to advancing this knowledge a teeny-weeny step.)

So it makes perfect sense to discuss color shifts (i.e., perceived color shifts) after adjusting L in the LAB color space.

Of course, the notion of “acceptable approximation” depends on context.  I trust the designers of XYZ and LAB that their mappings as good as it gets — but only in context of considering swatches of colors, not colors inside a real photo image. And when one discusses “contrast manipulation”, one is definitely out of context of color swatches, right?

See the OP: the tonal mapping people know it for a long time that when you change the contrast in the “brightness” channel, it is not enough to just naively preserve the original chromaticity (which would be fine in context of swatches!).  So considering color of swatches is just the first step. What I proposed in the OP is one such way to treat the next step after swatches.

Reply   Reply with quote   Complain
Antioch
Senior MemberPosts: 1,777
Like?
Re:Could somebody draw in the cat ! I do not see it !
In reply to ilza, Jul 10, 2013

ilza wrote:

Antioch wrote:

Toermalijn wrote:

Antioch wrote:

Regards

Both ralph and vicc pointed out the cat...

Yep, I saw both of the post pointing to it, but I still dont see it, as others have pointed out too. That is why I have asked for it to be painted out. At this point, I have to say, it is an imaginary cat ... lol.

Your feelings are very familiar to me (see the OP). However, after doing enhancements, and finally finding the beast, now I can’t NOT see it. Here is the central (yes!) crop of the enhanced image from the OP (cat is walking our way, or maybe ↘ way):

Thanks, without your painting it, I would never have seen it. Whew ...

Regards

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re: HSV works better than either RGB or LAB
In reply to Tom Axford, Jul 10, 2013

Tom Axford wrote:

For adjusting the tone curve, HSV works better than either RGB or LAB.

The V channel may be adjusted (even to extremes) without causing colour shifts and without the risk of channel clipping on conversion back to RGB.

My guts feelings are that HSL may have some advantages (comparing to HSV) in the situation when there is no clipping. And, of course, HSV is infinitely better in the cases when HSL clips. How to solve this?

If I would design a tonal-curve applicator, I would use a weighted average of two:

  • Apply the curve both to L of HSL and to V of HSV;
  • Calculate V of the converted HSL, take V of converted HSV (denote as V₁, V₂);
  • If V₁<0.5, keep it;
  • If V₁≥1, replace it by V₂;
  • For values of V₁ between 0.5 and 1, take the weighted average (with weights changing linearly between two boundary cases) V = tV₁ + (1-t)V₂;
  • t=2-2V₁

(This way, outside of zone IX, we use curve-on-L; in zone IX, we use something more complicated — but I expect it hardly matters what one does in zone IX.)

(There is no need to say that I would also “cook” saturation using the 45% rule.)

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re: Changing contrast without distorting colors: 45% rule and another cat
In reply to ilza, Jul 10, 2013

I wrote:

What you describe in this reference is, essentially, “the 100% rule”: you keep the saturation of the initial image intact. This is the “variant 2” of my post (with the variant 3 being the “the 0% rule”).

The whole topic of my post was that your method gives unnatural colors; hence “the 45% rule”. For my eyes, 45 gives visually better results than “the 40% rule” or “the 50% rule” — which are both significantly better than “the 100% rule” or “the 0% rule”.

To be more convincing, try “10 pictures are better than 20 words” approach.

  • Start with another popular “Find a cat” image (“Camouflage cat”; again, I could not find proper attribution);
  • Increase contrast ×5 (in GIMP, it is called γ=0.2: go to Colors/Levels, change the middle number);
  • Keep the original’s hue;
  • Keep a N% of the original’s saturation (N% rule; HOWTO-in-GIMP in the OP; changing is just moving one slider).
  • Compare the results for various values of N (below, 0, 40, 45, 50, 100).

The original (Camouflage cat)

γ=5, 0% rule (the variant 3 of the OP: apply curve to RGB, keep original’s hue)

γ=5, 40% rule

γ=5, 45% rule (for my eyes, the best match for the original’s colors)

γ=5, 50% rule

γ=5, 100% rule (variant 2 of the OP: keep hue and saturation of the original; this is what everybody else in this discussion thinks/thought is the best way to do it).

Judge for yourself… (BTW, I do not think that the contrast enhancement here helps with identifying a cat in any way. Did it help you?)

Reply   Reply with quote   Complain
Tom Axford
Senior MemberPosts: 1,148
Like?
Re: Changing contrast without distorting colors: 45% rule and another cat
In reply to ilza, Jul 10, 2013

Forgive me, but I am very slow and am having considerable difficulty in fully understanding your argument.

I think that the crux of the problem is understanding what you mean by your "N% rule". In your image labelled 100%, I can see that it has 100% of the original saturation. But your image labelled 0%, has more saturation to my eyes, not less, and it certainly isn't 0% saturation as that would imply a b&w image. I am obviously misunderstanding what you mean by the N% rule. Can you explain in more detail, please?

As to the realism of the various images, I agree with you that the most realistic images are those in between your 0% and 100%. To be honest, I would not care to make a judgement between your 40%, 45% and 50%, they are just too similar for me to be able to say which is most realistic.

However, I think it is a very interesting question that you raise and something I had not noticed before. However, I am not at all knowledgeable about this level of detail in theories of colour vision, and I rather doubt that there are many people on this forum who are experts in colour vision. I suggest that it may be better to find a more academic forum, populated by people doing research in colour vision, where you may find a more informed level of discussion.

Reply   Reply with quote   Complain
knickerhawk
Veteran MemberPosts: 3,156
Like?
Re: Changing contrast without distorting colors: 45% rule and another cat
In reply to Tom Axford, Jul 10, 2013

Tom Axford wrote:

Forgive me, but I am very slow and am having considerable difficulty in fully understanding your argument.

I'm pretty slow at this stuff too. To make things worse Ilza has marked me as a humanities major presumably incapable of understanding the difficulties involved. However, I'll muddle on as best I can...

I think that the crux of the problem is understanding what you mean by your "N% rule". In your image labelled 100%, I can see that it has 100% of the original saturation.

Yes, by N% he's talking about the % of the original saturation (pre-contrast adjustment) to be applied to the adjusted image.

But your image labelled 0%, has more saturation to my eyes, not less, and it certainly isn't 0% saturation as that would imply a b&w image.

The original image has relatively low saturation. The contrasty new version has relatively high saturation. 0% refers to the opacity setting not the actual amount of saturation. Thus, his 0% means 0% of the relatively low original saturation is being used and 100% of the relatively high saturation is being used. Conversely, 100% opacity means 100% of the relatively low saturation of the original image is being applied.

I am obviously misunderstanding what you mean by the N% rule. Can you explain in more detail, please?

As to the realism of the various images, I agree with you that the most realistic images are those in between your 0% and 100%. To be honest, I would not care to make a judgement between your 40%, 45% and 50%, they are just too similar for me to be able to say which is most realistic.

All the variants look like crap to me. But, hey, what do I know? I'm just a naive humanities major...

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re: Changing contrast without distorting colors: 45% rule and another cat
In reply to Tom Axford, Jul 11, 2013

Tom Axford wrote:

Forgive me, but I am very slow and am having considerable difficulty in fully understanding your argument.

I think that the crux of the problem is understanding what you mean by your "N% rule". In your image labelled 100%, I can see that it has 100% of the original saturation. But your image labelled 0%, has more saturation to my eyes, not less, and it certainly isn't 0% saturation as that would imply a b&w image. I am obviously misunderstanding what you mean by the N% rule. Can you explain in more detail, please?

As I said in the OP:

  • two strategies (variant 2 and variant 3) are marked as 100%-rule, and 0%-rule;
  • both strategies work with the original image, and the second image, which has the tonal curve applied separately to R,G,B channels;
  • both strategies keep the hue from the original image;
  • both take the brightness from the curve-applied image.

Now the variants differ in:

  • Variant 2: take the saturation from the original image (100% rule);
  • Variant 3: take the saturation from the curve-applied image (0% rule).

As you can see, the percentage reflects how much of the original saturation is preserved. And N%-rule just naively interpolates between two variants: it mixes N% of saturation of the original image with (100-N)% of saturation of the curve-applied image.

Is it more clear now?

However, I am not at all knowledgeable about this level of detail in theories of colour vision, and I rather doubt that there are many people on this forum who are experts in colour vision.

What I suspect is that the experts in color vision know close to nothing about color shift effects from the tone-mapping. (My judgement is based on one reference only — one I gave in the OP. It is one of the most influential papers on tone mapping, but what they propose to do with colors looks like a complete gobbledock to me.) I think that on this stage of research on this topic, what people who actually work with colors (and not just study them) say may have more weight than the opinions of academics.

–––––––––––––––––––––––––––––––––––––––––––––––––

Now: on differences between 40%, 45%, 50%. All of the images above are hideous; the middle ones are obviously less hideous than the 0% and 100% ones, but the general ugliness does not help in deciding which of 40%, 45%, 50% is better.  So maybe providing such an extreme case (γ=5) as an example was not my best judgement.

However, I tried this approach with “mild” curves (and more artistic images) many times.  45% was always inside the range of visually-the-best percentages.  This is why I say that for me it always work.  On the other hand, I have very limited experience with tonal curves; so I cannot realistically insist that my findings should better be followed by other people.

However, if “professional” retouchers voice their opinion here, this may eventually lead to more widely accepted (and probably better than what I propose) ways to treat saturation.  Then somebody might implement smarter tonal curves in their programs and/or academics may have bright unexpected insights on how all that may be looked-at/improved/reworked.

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re: Changing contrast without distorting colors: 45% rule and another cat
In reply to knickerhawk, Jul 11, 2013

knickerhawk wrote:

I'm pretty slow at this stuff too. To make things worse Ilza has marked me as a humanities major

Come on, all I said was that you behavedlike one. My apologies if you really took this simile too seriously!

presumably incapable of understanding the difficulties involved. However, I'll muddle on as best I can...

“Capability to understand” is a two way street. You not understanding me may be interpreted as your stupidity, or my ineptness, or just an impedance mismatch.

And I would say that for a humanities major, you muddle wonderfully!

On crappiness: IMO 45% is always better than 0% and 100%; but to find the best percentage more precisely, one needs either to consider a very large pool of images, or some extreme case, where the differences between 40%, 45%, 50% are more pronounced. I thought that γ=5 would provide a good example of such an extreme case; and to have something visible after γ=5, one must start with something sub-optimal.

Well, this was my logic in choosing this image as an example. In hindsight, it has obviously failed. Do you have better ideas for examples?

Reply   Reply with quote   Complain
knickerhawk
Veteran MemberPosts: 3,156
Like?
Re: Changing contrast without distorting colors: 45% rule and another cat
In reply to ilza, Jul 11, 2013

ilza wrote:

knickerhawk wrote:

I'm pretty slow at this stuff too. To make things worse Ilza has marked me as a humanities major

Come on, all I said was that you behavedlike one. My apologies if you really took this simile too seriously!

presumably incapable of understanding the difficulties involved. However, I'll muddle on as best I can...

“Capability to understand” is a two way street. You not understanding me may be interpreted as your stupidity, or my ineptness, or just an impedance mismatch.

And I would say that for a humanities major, you muddle wonderfully!

On crappiness: IMO 45% is always better than 0% and 100%; but to find the best percentage more precisely, one needs either to consider a very large pool of images, or some extreme case, where the differences between 40%, 45%, 50% are more pronounced. I thought that γ=5 would provide a good example of such an extreme case; and to have something visible after γ=5, one must start with something sub-optimal.

Well, this was my logic in choosing this image as an example. In hindsight, it has obviously failed. Do you have better ideas for examples?

I don't have Gimp installed and haven't used it, so I'm not sure what y=5 corresponds to in any scaling available in Photoshop. I'm also having difficulty understanding the problem you have with a straightforward curve adjustment set to blend in luminosity mode or a curve adjustment to the L channel in LAB. Depending on the contrast already present in the image, both work well without f'ing up color. Here's a quick and dirty luminosity blend of your image. Hue hasn't been changed by it. (The cat is sitting at the bottom of the rock pile a little more than a quarter of the way in from the right.)

Reply   Reply with quote   Complain
Tom Axford
Senior MemberPosts: 1,148
Like?
Re: Changing contrast without distorting colors: 45% rule and another cat
In reply to ilza, Jul 11, 2013

ilza wrote:

Tom Axford wrote:

Forgive me, but I am very slow and am having considerable difficulty in fully understanding your argument.

I think that the crux of the problem is understanding what you mean by your "N% rule". In your image labelled 100%, I can see that it has 100% of the original saturation. But your image labelled 0%, has more saturation to my eyes, not less, and it certainly isn't 0% saturation as that would imply a b&w image. I am obviously misunderstanding what you mean by the N% rule. Can you explain in more detail, please?

As I said in the OP:

  • two strategies (variant 2 and variant 3) are marked as 100%-rule, and 0%-rule;
  • both strategies work with the original image, and the second image, which has the tonal curve applied separately to R,G,B channels;
  • both strategies keep the hue from the original image;
  • both take the brightness from the curve-applied image.

Now the variants differ in:

  • Variant 2: take the saturation from the original image (100% rule);
  • Variant 3: take the saturation from the curve-applied image (0% rule).

As you can see, the percentage reflects how much of the original saturation is preserved. And N%-rule just naively interpolates between two variants: it mixes N% of saturation of the original image with (100-N)% of saturation of the curve-applied image.

Is it more clear now?

Yes, thank you. I think I understand it now. I will try out your method in GIMP on some of my own images and see what I think.

Unfortunately, going by what others have said, there is no equivalent method in Adobe software (PS, LR, etc.) so the great majority of readers of this forum will not be able to use it. That includes nearly all the professionals as I think they use PS almost exclusively (with one or two exceptions).

Anyway, although I'm strictly amateur, your method interests me and I will certainly give it a try over a period of time (not promising any quick response)!

However, I am not at all knowledgeable about this level of detail in theories of colour vision, and I rather doubt that there are many people on this forum who are experts in colour vision.

What I suspect is that the experts in color vision know close to nothing about color shift effects from the tone-mapping. (My judgement is based on one reference only — one I gave in the OP. It is one of the most influential papers on tone mapping, but what they propose to do with colors looks like a complete gobbledock to me.) I think that on this stage of research on this topic, what people who actually work with colors (and not just study them) say may have more weight than the opinions of academics.

I hope you will not be offended by a little personal advice: it does not encourage readers to take your work seriously when you are so dismissive about the work of others.

I have had a look at the paper you referenced and there is one short paragraph about handling saturation. They state that the method they use is essentially the same as previous researchers have used. Clearly their paper is primarily about their HDR method for transforming luminance and they have just picked up whatever was to hand to handle the colour.

If I understand their maths correctly, what they do to the saturation is simply take a percentage of the original saturation (arrived at by trial and error). That looks pretty crude, but it seems to work ok in the examples they show. They do not discuss how that method is arrived at, simply because they are borrowing it from previous research - which may give a justification (but I haven't looked).

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re: Changing contrast without distorting colors: 45% rule and another cat
In reply to Tom Axford, Jul 12, 2013

Tom Axford wrote:

I hope you will not be offended by a little personal advice: it does not encourage readers to take your work seriously when you are so dismissive about the work of others.

I have had a look at the paper you referenced and there is one short paragraph about handling saturation. They state that the method they use is essentially the same as previous researchers have used. Clearly their paper is primarily about their HDR method for transforming luminance and they have just picked up whatever was to hand to handle the colour.

This exactly MY point.  These people must be very competent, and thorough.  They must have investigated the subject, and picked the best way known.  And this way stinks.

If I understand their maths correctly, what they do to the saturation is simply take a percentage of the original saturation (arrived at by trial and error).

Yes, and they mutilate the saturation even in the parts of the image where contrast has not been changed!  And did not you notice that they manipulate the hue too — again, even where the brightness was not touched by their algorithm.

Do you still think that I’m “dismissive”?

Ilya

Reply   Reply with quote   Complain
ilza
Forum MemberPosts: 74
Like?
Re: Changing contrast without distorting colors: 45% rule and another cat
In reply to knickerhawk, Jul 12, 2013

knickerhawk wrote:

I don't have Gimp installed and haven't used it, so I'm not sure what y=5 corresponds to in any scaling available in Photoshop.

Exactly the same. Gamma is the one of the primary ways of image manipulation: it describes the contrast enhancement literally. If you have two pixels with levels differing by 1%, after applying γ=5, their levels will differ by 5%. (And γ=1 is DO-NOTHING.)

Googling, I can see that in Photoshop the UI for gamma is exactly the same as in GIMP: 1/γ is the middle of 3 number in the Levels dialogue. So just set this number to 0.2. (The formula for γ=5 transformation is newL=L⁵; here L is normalized to change between 0 and 1.)

I'm also having difficulty understanding the problem you have with a straightforward curve adjustment set to blend in luminosity mode or a curve adjustment to the L channel in LAB. Depending on the contrast already present in the image, both work well without f'ing up color.

It quite well may be that if you do very minor adjustments, and do not have “better tools” to compare your results to, the results will be satisfactory. Definitely, the colors will not be “f’ed up”. They just will be not the best one’s photoshop-fu can buy.

However, for printing, one may need to compress contrast ~1.5 times (γ=⅔). This already creates noticeable unnaturalness of colors (at least as my eyes can see it) if one uses the “touch only the brightness channel” method. With gamma still pretty close to 1, as above, the 45%-rule reduces these perceived color shifts to practically nothing.

With gamma very far from 1 (or very nonlinear tonal curves), using the naive ways such as N%-rule does not fix the color shifts problems completely (the colors in all the images I provided are hideous). Still, 45% image is, IMO, much closer to the initial colors than 100% one. Do not you agree?

Here's a quick and dirty luminosity blend of your image. Hue hasn't been changed by it. (The cat is sitting at the bottom of the rock pile a little more than a quarter of the way in from the right.)

Hue was not changed… This still has very little to do with the perceived change of colors. And compare the white wall on your image and on mine. Your manipulation is very mild. IMO, to be able to clearly recognize color shifts related to manipulating L, it is better to start with severe manipulations, where differences are more pronounced.

Ilya

Reply   Reply with quote   Complain
Tom Axford
Senior MemberPosts: 1,148
Like?
Re: Changing contrast without distorting colors: 45% rule and another cat
In reply to ilza, Jul 12, 2013

ilza wrote:

Tom Axford wrote:

I hope you will not be offended by a little personal advice: it does not encourage readers to take your work seriously when you are so dismissive about the work of others.

I have had a look at the paper you referenced and there is one short paragraph about handling saturation. They state that the method they use is essentially the same as previous researchers have used. Clearly their paper is primarily about their HDR method for transforming luminance and they have just picked up whatever was to hand to handle the colour.

This exactly MY point. These people must be very competent, and thorough. They must have investigated the subject, and picked the best way known. And this way stinks.

You are jumping to totally unjustified conclusions!

Why must they be competent and thorough? It is simply untrue to claim that every published research paper is written by people who are competent and thorough in everything they do. Even those papers that are very highly regarded and widely cited are not necessarily competent and thorough in every aspect of the work described.

Over my career I have read thousands of research papers (all in the distant past, I'm afraid), and the language and structure of this paper is very typical. My reading of it is that they have used a 'quick and dirty' method for handling colour that happens to give satisfactory results for the particular examples they use. They never claim otherwise.

If they were using the best known method, I would expect them to say so. Instead, colour is dismissed in a short paragraph in the section headed "Implementation'. The only justification they give for the method is that it gives 'satisfactory' results for their examples and that it was previously used by a couple of other people.

The topic of the paper is an HDR algorithm for use on the luminance channel only. It is ridiculous to expect it to be of the same high standard on matters that are purely peripheral. Research simply doesn't work that way. Researchers are very used to using quick and dirty methods on things that are subsidiary to the main thrust of their research. Frankly, they often wouldn't make any progress otherwise!

Reply   Reply with quote   Complain
Keyboard shortcuts:
FForum MMy threads