4

Why are the basis blocks corresponding to reflected waves in the quantisation matrix given seemingly random priorities in the standard JPEG quantisation matrices. Also, why aren't the priorities monotonic with respect to frequency?

I haven't been able to find any explanation and all I can come up with is possible tiling patterns occurring with symmetric quantisation matrices or an adaptation to the arrangement of photoreceptors in the eye.

  • Any given encoder can use whatever quantization values it wants. I have seen quantization matrices that are both monotonic and symmetrical and those that are not. The quantization matrices are, ideally, values chosen to give a good trade off between compression and appearance. As some point, someone has probably lined up a bunch of images, compressed and decompressed, with various quantization tables and asked a group of people, "Which looks best?" That test depends upon the type of image used. There many applications out there for finding optimal quantization tables for a specific image. – user3344003 May 05 '14 at 17:09
  • 2
    The classic book "JPEG: Still Image Data Compression Standard" by Pennebaker and Mitchell discusses how the JPEG committee developed the standard quantization parameters. The values were picked trying to minimize values while maintaining a certain image quality. My dim memory is that they used an A-B experiment to determine when people saw image degradation. – garlon4 May 06 '14 at 18:45

1 Answers1

2

The quantization tables are a set of fudge factors that attempt to model human perception.

The specific quantization table values are more art than science, because human perception is quirky and complex, and ideal coefficients depend on specific viewing conditions that can only be roughly guessed in advance.

Tables are not always monotonic with respect to frequency, because blocks of certain frequencies form patterns that are more useful than others, e.g. for straight horizontal and vertical lines.

Kornel
  • 97,764
  • 37
  • 219
  • 309