High performance arithmetic coding for small alphabets

Abstract

Summary form only given. Generally, there are two main obstacles in the application of arithmetic coding. One is the relatively heavy computational burden in the coding part, since at least two multiplications are needed for each symbol. The other is that a highly efficient statistical model is hard to implement. We observe that under some important circumstances the number of different symbols in the data stream is definitely small. We specially design both the coding part and the modeling part to get a high performance arithmetic coder for the case of small alphabets. Our method is based on the improved arithmetic coding algorithm. We further improve it to be multiplication-free.