Research on Practical Implementation of Binary Arithmetic Coding with Forbidden Symbol for Error Resilience

Arithmetic coding is a widely applied compression tool with superior coding efficiency to other entropy coding methods. However, it suffers from the error resilience and complexity. In this paper, the integer implementation of binary arithmetic coding with forbidden symbol for error resilience is studied. Coding redundancies for employing different quantization coefficients in probability representation and cost effective backtracking distance in bits for maximum a posteriori (MAP) decoding are studied in depth. We observe that the optimal quantization coefficients are independent of forbidden symbol and the probabilities of source and the cost effective backtracking distance is related to the source entropy and the given forbidden symbol probabilities. Such observations are also demonstrated by extensive experiments.