Create quantization matrix

Pls explain your context. quantization matrix ... for what? If you're dealing with JPEG image compression (image blocks + DCT quantization + quantization + huffman), the compressor has the freedom to use its own quantization matrix - or rather a family of matrices, one for each quality factor. "

Obviously, you usually want to assign a lot of bits to the low frequency components, and a few to the high frequencies, but this applies to everything that can be said in general.

Also, keep in mind that JPEG compresses luma and chroma (and chroma is usually a sub-selection), so different matrices can be used for each.

I believe the standard offers some typical matrix, eg including scaling factors for different qualities. But this is not required at all. Also, you can find (googling!) Many matrices here for many cameras and graphics applications.

Update: From here :

Setting up quantization tables for the best results is a black art and an active area of ​​research. Most existing encoders use simple linear scaling of the sample tables provided in the JPEG standard using one user-defined "quality" to determine the scaling factor. This works well enough for midrange qualities (not too far from the samples themselves), but rather suboptimal at very high or low quality settings.