Abstract: In order to solve the more serious context dilution problem faced by the modeling of M-ary source sequences than the modeling of source sequences,a lossless compression algorithm based on context tree model is proposed. First, the context information of the symbols to be coded and the principle of reducing entropy value in the information theory are used to establish a context tree model to make the statistical information of the source more detailed and convert the M-ary tree to a binary tree. Then, in order to obtain a conditional probability distribution that is more conducive to improving coding performance, the increment of the description length is introduced for the merging of tree nodes. Finally, in order to deal with the problem of zero-probability symbols generated during the arithmetic coding process, the escape symbol with non-zero frequency is introduced into each conditional probability distribution, which improves the disadvantages of traditional methods. The experimental results show that the proposed algorithm can achieve better compression results.