Now, two additional things are done:

- The value of
*k*does not have to be fixed. It can also be adaptive, as long as the encoder and decoder always make the same choice of*k*. We increase*k*after emitting a 0 bit and decrease it after emitting a 1 bit. - Secondly, we make a distinction between significant bits and so-called refinement bits. Significant bits are those whose most significant bit has
*not*already been sent / coded. Consequently, refinement bits are the ones for which more significant 1 bits have already been coded. The significant bits have a high probability of being 0, the refinement bits in contrast only have a probability close to 0.5 of being 0. Thus, we ignore the refinement bits when run-length coding zero-runs of the significant bits (as the decoder is aware for which coefficients this is the case) and embed the "ignored" refinement bits into the bit-stream without any coding after writing the next 1 bit.

- http://en.wikipedia.org/wiki/Rice_coding
- Henrique Malvar, "Progressive Wavelet Coding of Images", 1999, IEEE Data Compression Conference March 1999

Generated on Tue Jul 10 20:44:34 2007 for wavelet by 1.5.2