Compression Discussion

[NA9D]

Re: User-requested enhancements & features?

After compression and decompression, the blockiness really suggests that pixels are affected by their neighbors. The closer the neighbor is to your value, the more likely you are to lose your unique identity, which implies a loss of information.

This is 100% correct. When I was writing compression routines for still images back in 1988 when I was in Japan, this was what the basis for our whole algorithm was. You took a pixel and compared its value to its neighbors. If it was close, you could effectively eliminate it. The quality of the resulting final image could change dramatically depending on how large you allowed that "delta" between the images to be.

[NA9D]

Re: User-requested enhancements & features?

I'm sorry to disagree with you here, but if you compress something and it doesn't come back identical to what you started with, you have lost information. I'm not sure why you are being so critical of this point. I am 100% on target with what I said here. I chose to say it in a very understandable way to differentiate between "guessing" what the pixels might be (for upconverting DVDs) vs. losing the correct value for pixels that are less relevant to the overall picture. I agree with Hodge, you are sounding a bit grumpy here.

Why I am being specific in saying that there are no "lost" pixels has a purpose. The poster who kinda started this discussion suggested that Vudu's HD content might not really be HD. In reality that would mean that the pixel count of the content is not 1024 x 768 or higher or as in Vudu's case 1920 x 1280. The fact is that there is always 1920 x 1280 pixels. I felt that your statement of "losing" pixels indicates that maybe the initial 2,457,600 pixels is some number less than that. My point is that the resolution in pixels of the picture does not degrade. The quality of the image degrades but the pixel count stays the same.

I'm debating semantics with you!

[redwein]

Re: User-requested enhancements & features?

BTW redwein, I've had a nice chit-chat with you tonight. Thanks!!!

Ditto.

[redwein]

Re: User-requested enhancements & features?

Your point is valid, but I don't like the answer. It sounds like doom and gloom and there is no answer to the question.

I don't think of it as doom and gloom. I foresee a future where there is enough bandwidth to send BluRay quality data over the network. Once that happens the point is moot.

[HeadHodge]

Re: User-requested enhancements & features?

BTW redwein, I've had a nice chit-chat with you tonight. Thanks!!!

[HeadHodge]

Re: User-requested enhancements & features?

With enough data, I'm sure you could get an average. I still question its value though. A number really doesn't tell you how you will perceive something like that. I'm sure higher is better but it's possible that you could have a measured difference that is less than the threshold of human detection.

Your point is valid, but I don't like the answer. It sounds like doom and gloom and there is no answer to the question.

I read on one of the posts awhile back that there is a technique to capture and reapply the missing info to a frame to essentially recreate the original. I can't remember who posted it or how it works, but if it's true it would suggest that there is a way to quantify the amount of loss caused by compression.

[redwein]

Re: User-requested enhancements & features?

With enough data, I'm sure you could get an average. I still question its value though. A number really doesn't tell you how you will perceive something like that. I'm sure higher is better but it's possible that you could have a measured difference that is less than the threshold of human detection.

[HeadHodge]

Re: User-requested enhancements & features?

I think you are wrong. Just look what happens with the blockiness of a low detail image. That seems to suggest that there is a bigger delta than a high detail scene where you can read fine print in a newspaper. In an uncompressed format, both scenes would take exactly the same amount of information since each pixel would be independent. After compression and decompression, the blockiness really suggests that pixels are affected by their neighbors. The closer the neighbor is to your value, the more likely you are to lose your unique identity, which implies a loss of information.

Ok, let's say I'm wrong (although I can't prove that either), then let's take and compare all 120,000 frames and take an average of the percent difference as the metric. I'm already guessing your going to say that each movie is going to be different depending on how much action or darkness in it.

But even if my proposal is poo-poo, you're a smart guy there must be a metric that can be conceived that's better than what we have today.

[redwein]

Re: User-requested enhancements & features?

Well the basis of my whole theory, is that a VBR compression is exactly that, variable bit rate compression, but the end result is a consistent loss of quality on a frame by frame basis.

I think I'm right, but I don't really have a way to prove it.

I think you are wrong. Just look what happens with the blockiness of a low detail image. That seems to suggest that there is a bigger delta than a high detail scene where you can read fine print in a newspaper. In an uncompressed format, both scenes would take exactly the same amount of information since each pixel would be independent. After compression and decompression, the blockiness really suggests that pixels are affected by their neighbors. The closer the neighbor is to your value, the more likely you are to lose your unique identity, which implies a loss of information.

[HeadHodge]

Re: User-requested enhancements & features?

But what's the point of that? You could pick a particular frame that may be 99% identical and another is 75% identical. I don't think it's possible to have a meaningful metric on such a moving target. The selection of the frame that you are suggesting is an entirely arbitrary choice. At 24 fps, there are over 120,000 frames in a 90 minute movie.

Well the basis of my whole theory, is that a VBR compression is exactly that, variable bit rate compression, but the end result is a consistent loss of quality on a frame by frame basis.

I think I'm right, but I don't really have a way to prove it.

[redwein]

Re: User-requested enhancements & features?

Yes what you say is true. But what I'm talking about is comparing a single frame at the source to the same single frame at the destination. Either they are identical or there not.

But what's the point of that? You could pick a particular frame that may be 99% identical and another is 75% identical. I don't think it's possible to have a meaningful metric on such a moving target. The selection of the frame that you are suggesting is an entirely arbitrary choice. At 24 fps, there are over 120,000 frames in a 90 minute movie.

[HeadHodge]

Re: User-requested enhancements & features?

That's not so easy to do since the delta (number of different pixels) will vary wildly from one frame to another due to differences in the content. Just look what happens with a digital camera when you take pictures. The camera captures the same number of pixels but the file sizes vary dramatically.

Yes what you say is true. But what I'm talking about is comparing a single frame at the source to the same single frame at the destination. Either they are identical or there not.

[redwein]

Re: User-requested enhancements & features?

I just would like to see a metric that you can use for comparing the original to the end result.

That's not so easy to do since the delta (number of different pixels) will vary wildly from one frame to another due to differences in the content. Just look what happens with a digital camera when you take pictures. The camera captures the same number of pixels but the file sizes vary dramatically.

[HeadHodge]

Re: User-requested enhancements & features?

No. It's a very good day.

I just don't know how many times I must repeat myself about this HD quality stuff!

I am tired and I've had 3 raspberry martinis tonight....

So you're drinking martinis and you're only giving me meds?? What a rip-off!!

The only point I've been trying to make today, is that if you take one frame of video that the studio provides VUDU and you compare that to the exact same frame that ends up on the VUDU box, wouldn't you agree that they are probably not identical???

So my proposition for a measurable metric of quality is simply the comparision of those two frames. If they are identical then it's 100% match, if they are not then it's some percentage thereof.

So I don't care how the frame is transmitted to the box, or even if it's a 100% match. I just would like to see a metric that you can use for comparing the original to the end result.

[redwein]

Re: User-requested enhancements & features?

I am tired and I've had 3 raspberry martinis tonight....

That explains it.