Complexity Analysis of Adaptive Binary Arithmetic Coding Software Implementations

This paper is dedicated to the complexity comparison of adaptive binary arithmetic coding integer software implementations. Firstly, for binary memoryless sources with known probability distribution, we prove that encoding time for arithmetic encoder is a linear function of a number of input binary symbols and source entropy. Secondly, we show that the byte-oriented renormalization allows to decrease encoding time up to 40% in comparison with bit-oriented renormalization. Finally, we study influence of probability estimation algorithm for encoding time and show that probability estimation algorithm using “Virtual Sliding Window“ has less computation complexity than state machine based probability estimation algorithm from H.264/AVC standard.