Improving visual quality for lossy coding

[palemieux]

The only parameter available to ojph_compress to control lossy quality is -qstep, which AFAIK automatically generates quantization step sizes in each sub-band and optimizes PSNR. The latter results in very noticeable visual artifacts at lower qualities.

For example, the leftmost tile was encoded at with -qstep 0.025. For comparison, the center tile was encoded using KDU with -rate 0.4 Ctype=Y and the rightmost tile with Qfactor=86. Note that Qfactor removes some but not all the high-frequency artifacts. All encoded images were around 466 kB.

Ideally, OpenJPH would ultimately include visual weighting options at encoding. In the meantime, it might make sense to note limitations at low quality/bit rates.

Source image (thanks @hmaarrfk):

rev.zip

[aous72]

Hi Pierre,

Thank you for raising this issue.
I totally agree that I need to add visual weighting and quality factor; it is just that I have not had the time to do them.

Kind regards,
Aous.

[palemieux]

@aous72 Based on the above, it is not clear if Qfactor will be sufficient. What are your thoughts?