Hi Timo Kokkonen,
I'd like to suggest a new option to jpegoptim: the adaptive (dynamic) Huffman coding.
After reading this paper:
it turns out that the usual custom Huffman table (also called optimized JPEG) yields on average a reduction of 1.38% of image's size, but the use of dynamic Huffman coding (which builds an even better Huffman table) almost doubles this figure by yielding on average a further reduction of 1.01%.
An implementation of Vitter's algorithm (an improved version of the algorithm mentioned in the previous paper and which yields even better reductions) is released in the public domain and can be found here :
Given all this, would it be possible to add this new option to jpegoptim to optimize baseline and progressive JPEGs?