I am now looking at implementing a fast Huffman encoding for the RAW data in a DNG container. I looked at the code inside DNG SDK as well as other samples around the web. All the samples use 8-bit data. I can unpack 12-bits into 16-bits, and then use the first frame to compute the probabilities. Can someone point me to huffman implementations optimized for RAW files ? I am not an expert at this, but the DNG SDK did not seem to be very optimized. lz4 compression is an excellent example of optimized implementation and coding.
Such as ?
You can do some preprocessing before Huffman.
There is no way of using temporal dependancy ?
I mean that Huffman encoding is just compression technique, if it is done right it does not matter if it is used for raw compression or other things.
DNG only supports Huffman encoding for RAW images. Hence my interest in Huffman.
Huffman compression has nothing to do with raw. You need to just find good implementation. Check some Intel libraries or such.
It looks like you're new here. If you want to get involved, click one of these buttons!