psychovisual benchmarks?

[tetsuo55]

Hey everyone.

I think most people try to reach the following goal when transcoding and using avisynth scripts.

High psychovisual image quality at an as low as possible bitrate.

From what i understand SIMM and PSNR benchmarks look at physical difference between both images before and after. What i am looking for is a benchmark that takes psychovisual data into account.

That would mean that the SIMM could be very low but the psychovisual-benchmark result could be very high

If we had such a benchmark we could possible achieve higher compressions rates at the same percieved image quality (then again i could be totally wrong)

[unskinnyboy]

What tool(s) would you use to quantify perceived quality?

[tetsuo55]

My guess is that a new tool would have to written based on current research results

[unskinnyboy]

What current research results?

Anyway, the point I was trying to arrive at was that, it's not possible (at least, at this time) to mathematically quantify what you and me perceive, unless we have wires embedded in our brain or have real-time brain scans while viewing the image/video.

[tetsuo55]

Ahh thats not correct.

There is a lot of research done on the psychovisual of the human eyes and brain, just like the research done to create mp3.

Those results can be translated to mathematical data (things like on average humans cannot see detail above XXX or see colors above threshold XXX)

Some documents about the research (not all usefull to this cause)

http://labs.divx.com/node/90
http://citeseer.ist.psu.edu/588567.html
http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel1/2220/10462/00490978.pdf?arnumber=490978 << looks like you have to pay for that one :(


google has a load of research documentation about this, also document referring explicitly to quantization

[unskinnyboy]

The links you pointed to, don't all that talk about how to apply PE, different algorithms and ways to do it etc. (I didn't go through all the papers in the second link since there was a ton of them, so if there's some paper I should specifically look at, let me know)? That's all fine and dandy, but I still fail to see how an Avisynth filter can be used to define how I interpret an image inside my brain.

[tetsuo55]

ill try and find more relevant documentation, but you could be right that that avisynth might not be the right place for this