Binarization is an elementary step in the framework of context-based adaptive binary arithmetic coding. It is introduced to simplify the context modeling and the subsequent arithmetic coding process. In this paper

we propose a novel binarization scheme for arithmetic coding. Based on the observation that the numbers of nonzero coefficients in neighboring blocks are highly correlated

the number of nonzero coefficients of a block is specifically binarized according to the numbers of nonzero coefficients of its top and left neighbors. In addition

a novel mapping rule is designed to map a syntax element to a code number

according to the observed features of probability distribution. Experimental results show that compared to the conventional scheme

this scheme can reduce decoding complexity with comparable coding efficiency.