CS631 Paper Response Form

Paper Title: Experiments with a Lossless JPEG Codec

Author(s): Kongji Huang

Main Point(s):

An empirical evaluation of the different prediction selection values of the lossless JPEG standard shows that in general, the two-dimensional predictors perform better, and that optimal performance cannot be achieved by a single predictor but rather by dynamically choosing the most appropriate predictor. Comparison with other compression techniques show that lossless JPEG gives very good compression ratios on a wide range of images, but is somewhat slower.

Possible Use(s):

These results helps one to trade off time/compression for image compression.

Extensions:

It seems like the algorithm chooses one PSV for the whole image. I presume one could do better by chopping up the image in smaller units and choosing PSV's that are optimal for these blocks. Experiments would have to show what would be the optimal size of these blocks, because the gains have to set off the overhead of sending an additional 3 bits per block for encoding the chosen PSV.

CS631 home page