How are the values of the quantization matrix determined?
tap22sf
I suspect a lot of these techniques could be applied to real time 3D rendering as well. I am not aware of this as a common technique today, but reducing the rate of shading based on maybe texture or geometric complexity could be cool. Sample higher on edges.
ecohen2
Is there a threshold for when the numbers are "eliminated" is it just working with a floor function?
azwarens
Looks like the quantization matrix is fairly standardized - the one on the slide is referred to as the common one in a few articles.
"Two examples of quantization tables are given in Tables K.1 and K.2. These are based on psychovisual thresholding and are derived empirically using luminance and chrominance and 2:1 horizontal subsampling."
"At this point, one might wonder how the values in the quantization matrix are selected. An enormous number of schemes could be used to define these values, and the two most common experimental approaches for testing the effectiveness of such schemes are as follows:
Measure the mathematical error found between an input image and its output image after it has been decompressed, trying to determine an acceptable level of error.
Simply "eyeball it". Although judge the effect of decompression on the human eye is purely subjective, it may, in some cases, be more credible than mathematical differences in error levels.
Although JPEG allows for the use of any quantization matrix, ISO has done extensive testing and developed a standard set of quantization values that cause impressive degrees of compression."
Is there a clearer way to represent "per-element divide?" I kept trying to figure out what was going on, and then realized we weren't doing "right-side divide by quantization matrix." (Maybe a note saying that the divide is per element?)
maq
How does the division work? Not really sure where the 0s are coming from, since many of the element-wise divisions do not seem to actually end up to 0. Is there some kind of rounding going on in addition to the element-wise division?
How are the values of the quantization matrix determined?
I suspect a lot of these techniques could be applied to real time 3D rendering as well. I am not aware of this as a common technique today, but reducing the rate of shading based on maybe texture or geometric complexity could be cool. Sample higher on edges.
Is there a threshold for when the numbers are "eliminated" is it just working with a floor function?
Looks like the quantization matrix is fairly standardized - the one on the slide is referred to as the common one in a few articles.
This site has quantization matrices by digital camera: https://www.impulseadventure.com/photo/jpeg-quantization.html
"Two examples of quantization tables are given in Tables K.1 and K.2. These are based on psychovisual thresholding and are derived empirically using luminance and chrominance and 2:1 horizontal subsampling."
Source: JPEG Standard, Annex K (see page 147 of the PDF).
Another post on JPEG compression states:
"At this point, one might wonder how the values in the quantization matrix are selected. An enormous number of schemes could be used to define these values, and the two most common experimental approaches for testing the effectiveness of such schemes are as follows:
Although JPEG allows for the use of any quantization matrix, ISO has done extensive testing and developed a standard set of quantization values that cause impressive degrees of compression."
Source: https://cs.stanford.edu/people/eroberts/courses/soco/projects/data-compression/lossy/jpeg/coeff.htm
Is there a clearer way to represent "per-element divide?" I kept trying to figure out what was going on, and then realized we weren't doing "right-side divide by quantization matrix." (Maybe a note saying that the divide is per element?)
How does the division work? Not really sure where the 0s are coming from, since many of the element-wise divisions do not seem to actually end up to 0. Is there some kind of rounding going on in addition to the element-wise division?