Sharpness Metrics search

[gelo333]

Are there software video sharpness meters? When the sharpening filters analyze the video before the action, they get some calculation. Why can't this calculation be displayed as an indicator for the user?

Why is this needed. When trying various filters, you need to work with your eyes, this overloads your eyesight. Also, your vision can be different every day. Today you feel like this, tomorrow it may look different. This is a very long process in the end.

[Cornucopia]

Sharpness is perceptual, meaning it is realized in your mind.

Sharpness is a function of resolution, global/overall contrast, local/edge contrast, motion (or lack), content complexity, personal optical sensitivities/thresholds.
How do you expect a filter to calculate that?


Scott

[gelo333]

There are such methods. But personally I do not know how to apply them, I do not have such knowledge.

For example

https://pypi.org/project/cpbd/
https://blog.csdn.net/dcrmg/article/details/53543341

[Cornucopia]

The ONLY arbiter of whether your processing is good enough or not IS your eyes. It all comes back to that.

I have seen plenty of threads here where someone will ask for help, get a suggestion, try it and be very happy while others on the thread see the result and can immediately tell that the result was OVERSHARPENED with resulting haloes, etc. And also threads where it worked the other way, people think a process looks good but the OP is not satisfied.
Those methods you mention do NOT account for that, they just give some statistical average. But perception isn't a single figure, it is a spectrum. Do YOU fall into that average? Do I? Nobody knows until you verify it again going back to your own perception.

Scott

[gelo333]

I will suggest a different situation.

Do you have the original video. Then you carry out optimizations and changes with the file in various ways, option-1 and option-2. How to understand which option is sharper without looking at it for a long time? One calculation will be higher, another calculation will be lower. And so I can make a choice.

[Cornucopia]

What's all this long time stuff?

Look at your footage in a few select points which would be more challenging, then make your choice.

Scott

[_Al_]

There are such methods. But personally I do not know how to apply them, I do not have such knowledge.

For example

https://pypi.org/project/cpbd/
https://blog.csdn.net/dcrmg/article/details/53543341

To sum up that cpbd link. It is in python. You'd need to install latest Python on your pc.
Then for cpbd you'd need to instal that cpbd python module: pip install cpbd (in cmd window) and also scipy python module
Then you need to run python script that is on that pypi page. But that needs bmp, so you'd need to extract a bmp from your video frame.
Then you'd get a number. So you'd need sort of reference what number means what sharpness based on your experience and resolution and perhaps even categorized for same kind of video.
But for most likely usage scenario you'd need to run it with your original video image and then comparing it with your processed video image, meaning comparing those cpbd numbers. How much usable that would be is checking it for yourself, getting some affordable difference ranges etc.
This is all just my wild guess for an approach. Could be wrong though.

That other link is in C language and opencv.

On this forum, there are people knowledgeable in Avisynth or Vapoursynth and there might be some finesse to come up what is sharper. Specifically it might not be a problem to come up if it is sharper than original and come up with some sort of scale how much. If you masks of contours, edges and such or other methods. I did not google for any scripts, some scripts might be available. In Avisynth you might have more methods ready and such, vapoursynth maybe less filters (but both have Sobel basics and similar), also vapoursynth is python so final comparisons and deliveries are a bit easier.

[gelo333]

What's all this long time stuff?

Look at your footage in a few select points which would be more challenging, then make your choice.

Scott

this is the idea of automatic metrics. eyes see differently, your mood changes, the technique shows in different ways. it is not an objective result. otherwise there would not be many image analysis software tools.

[gelo333]

...

I asked people who know Python. They said cpbd won't work because it's old and hasn't been updated.

I'm sure that filter developers will be able to create a sharpness-metric tool. But I was search for a ready tool. And why is there no such tool yet?

[lollo]

You started with

Are there software video sharpness meters?

Then you added references for the tools

https://pypi.org/project/cpbd/
https://blog.csdn.net/dcrmg/article/details/53543341

and wrote you are not able to use them.

It has been said to you that the only "good" method is your eyes (and I agree). So, what are you finally trying to prove? That using own eyes is not a good method? Or are you looking for somebody teaching you how to use the tools you mentioned or other tools?

There is nothing wrong using own eyes to judge the results of the various AviSynth/VapourSynth filters. Here every experienced user do this to death without issues.

[gelo333]

Then you added references for the tools

These are methods for pictures, not for videos. And one of them is not working. Evaluating results with the eyes is not at all a scientific method.

Or are you looking for somebody teaching you how to use the tools you mentioned or other tools?

I wanted to find ready-made metric methods.