A fast renormalisation for arithmetic coding

Abstract

Summary form only given. All integer based arithmetic coding consists of two steps: proportional range restriction and range expansion (renormalisation). Here a method is presented that significantly reduces the complexity of renormalisation, allowing a speedup of arithmetic coding by a factor of up to 2. The main idea is to treat the output not as a binary number, but as a base 256 (or other) number. This requires less renormalisation and no bitwise operations.