Context Quantization Algorithm Based On Description Length

In the system that uses the Context model to implement the compression coding,The Context model is an effective method to estimate the source probability model.And the focus of research on Context modeling is to improve the compression efficiency by Context quantization.The essence of Context quantization is to reduce the number of conditional probability distributions in the model.In this paper,multi-dimensional Context quantization algorithm based on dynamic programming,hybrid clustering algorithm based on K-means and the algorithm combined unsupervised extreme learning and K-means clustering are proposed for achieving the quantification of Context mode.Then the quantitative performance of the three algorithms is compared and analyzed,and the relative optimal Context quantization algorithm is determined.In addition,because encoding efficiency of the encoding system is determined by quantitative performance and label cost,so in the light of hybrid clustering algorithm based on K-means and the algorithm combined unsupervised extreme learning and K-means clustering,label cost which generated in the process of quantification also is analyzed.Because of a large amount of computation,in this paper,the multi-dimensional Context quantization algorithm based on dynamic programming is used to achieve the quantification of single-condition and two-condition Context models;hybrid clustering algorithm based on K-means and the algorithm combined unsupervised extreme learning and K-means clustering are used to achieve the quantification of single-condition to sixcondition Context model.Among them,the description length is mainly used as an evaluation criterion for quantifying performance.Thus,the optimal condition number of the model,a better Context mode can be provided.By analyzing the experimental results,the quantification performance of the Context model of the above three algorithms is verified.The comparison of the experimental results shows that the hybrid clustering algorithm based on K-means can make the Context quantization performance better.In addition,in the light of the hybrid clustering algorithm based on k-means and the algorithm combined unsupervised extreme learning and K-means clustering,quantization performance of six-condition Context model is better than quantitative performance of other Context model.Other than,if considering the whole encoding system,namely the label cost,we can get: not only will the label cost affect the coding efficiency,but also affect the choice of the optimal Context model and the context quantization algorithm.