Local Tone Mapping (LTM) quality evaluation

[KronoX]

Good morning,

I am kinda new in this area but I find it really interesting.

I am doing a reasearch on evaluating how to improve the LTM method.
However, I find out that the criteria for indicating if one image is better
than other can turn really subjective.

As I studying the tutorials I observed that in the following link:
https://www.cambridgeincolour.com/tu...ty-filters.htm
in section "Move Location of GND Filter", there is an image in which
"Moving Location of GND Filter" can be done. One of the buttons
to do so says "Optimal"(and of course, it really seems to be the
optimal one). But, I was wondering if there is any scientific papers
in which I could investigate how to get an OPTIMAL image
I know that was no LTM method, but felt free to relate the OPTIMAL
concept associated.

To make an objective evaluation I am performing: sharpness, luma noise,
texture tests (actually I am looking on how to evaluate chroma noise also).

Thank you in advance,

[Mike Buckley]

To make an objective evaluation I am performing: sharpness, luma noise,
texture tests (actually I am looking on how to evaluate chroma noise also).

Those characteristics are also subject to completely subjective opinion. There is nothing about having made a photographic image that is not subject to subjective opinion about whether or not a particular characteristic is an attribute.

[KronoX]

Thank you Mike. Actually, I was aware of that but was hoping there was some thing I could take as reference. Anyway, I was thinking on showing pictures side by side to a group of people so they can give me a feed back.

Again, thank you very much.

[Manfred M]

Welcome to CiC. If you wouldn't mind clicking on the "My Profile" button on the top of this page and entering at least your first and and where you are from? We are a fairly informal group and go about on a first name basis here.

Let me start off by challenging your premise of "OPTIMAL image". Simply put that does not exist either from a compositional (artistic) or technical (related to the function of the camera and / or camera settings). Modern digital cameras, especially once you get into the higher end amateur or professional ones and use a modern lens you can get a really decent image that shows barely perceptible shadow detail as well as highlights that show some degree of tonality. These cameras are easily getting 12 eV or exposure with some of the top ones approaching 15 eV (at base ISO). Unless you are shooting in some very specific circumstances (sunrise or sunsets, shooting in a shaded forest where you see some bright light poking through the trees) the photographer does not usually have to worry about the dynamic range in much of his or her work).

To me this defines the "optimal" image, i.e. my camera has captured all the data I need to craft the image that I want. To try to define other optimizations (low noise, for instance) are non-starters. Although noise, especially chroma noise can be viewed as problematic in many images, it can also be used as a compositional tool in others. Film grain was very much viewed this way in the film days.

Tone mapping can be useful if the lighting in the scene exceeds the dynamic range in the image, but so do other methods of image blending. I tend to handle the blending of multiple images manually, but there are some automated solutions outside of the classical HDR programs; Enfuse comes to mind.

http://enblend.sourceforge.net/).

Your questions on the GND filters is an interesting one and one that has many opinions wrapped around it, as for the most part the effects can be accomplished by either using a physical filter or by applying a gradient in post production. The only time that I find shooting with a grad is a "must" is in situations where the sky is so bright that parts of it will clip (although bracketing the shots and blending in post are also an option).

So to answer your initial question, there is no definition of "optimal", as it does vary from image to image and from user to user. Getting clean, unclipped data in the raw file is about as close to the definition that I am willing to go.

[Manfred M]

Anyway, I was thinking on showing pictures side by side to a group of people so they can give me a feed back.

That is likely going to provide you with meaningless results. Unless you can isolate specific aspects of an image cleanly and have a statistically valid sample size, your results will be meaningless. I would suggest that showing the same image in different settings (say European, Asian and North American) will get you different answers.

I suspect you are chasing a question that can't be answered.

[Mike Buckley]

I was thinking on showing pictures side by side to a group of people so they can give me a feed back.

You can do that here at CiC. Even so, the best reason to do that for my way of thinking has not as much to with obtaining feedback about which of two photos is better as much as generating discussion about the process of deciding why one image is better. We can all learn to improve our process about making that decision. However, in the end, each of us has to decide for ourselves what we like and why we like it.

The one exception to that is when our primary goal is to make an image that pleases someone else, such as a paying client or the recipient of a photo we are giving as a gift. Even then, we're making the photo a certain way because it's more important to us that the other person likes the image more than that we like it.

[KronoX]

Hi GrumpyDiver,

I already updated my profile data.
Thank you for the time you took to answer. I found interesting your reply, and examples.

Best regards.

[KronoX]

Once again, thank you Mike.

I didn't see your last reply before.

[xpatUSA]

I am doing a reasearch on evaluating how to improve the LTM method. . . . I was wondering if there is any scientific papers in which I could investigate how to get an OPTIMAL image

Buon dia Ricardo,

Have you seen these?

http://www.disneyresearch.com/wp-con...-Pub-Paper.pdf

http://www.google.com/patents/US20120170843

http://www.eurasip.org/Proceedings/E...1569583991.pdf

(you did say scientific )

To make an objective evaluation I am performing: sharpness, luma noise,
texture tests

If you would like to download RawTherapee, it has several different methods of increasing local contrast and it also has deconvolution sharpening. The main reason that I tell you this is that each saved file has also a sidecar file which can be viewed in Windows Notebook and all of the contrast and sharpness setting are recorded there.

. . . actually I am looking on how to evaluate chroma noise . . . .

First you must find a method to convert an RGB file to an LCH file (lightness, chroma, hue) and then a method to view only the chroma part. Then use a program like ImageJ to analyze any area in the chroma image to obtain the noise standard deviation which is the evaluation you are looking for, if I understand correctly.

If that sounds difficult, it might be easier to find something that converts RGB to HSL space and then to analyze the saturation part. Saturation in HSL space is similar to chroma and could be used for the purposes of comparison.

[KronoX]

Buenos días / Bom dia Ted,
(Hahahaha actually I'm leaving at Brasil but I am peruvian, i think that will explain why I answered also in spanish )

Thanks a lot for your answers,

Actually:
- I read that Disney research, but as it is for videos (uses temporal noise reductors) it cannot be applied to what I am looking for;
- the Adaptative LTM of Color Images is also into my to-read list, but I really appreciate it; and
- the other paper named "Methods for LTM", is one I didn´t find before, I think it can help me.

I have a method that performs LTM, but I want(ed) it to achieve an "optimal". I will try to implement
the methods of the last paper in order to compare it results to mine, and perceive what I sall modify.

Cheers,

Ricardo