The Application Of Context Quantization Algorithm Based On Description Length In Wavelet Image Compression

Along with the development of an era, science and technology have integrated into every corner of life. Our production, living, and entertainment are depending on and expanding the digital world. Images and videos have the largest quantity among all kind of the digital data. Entropy coding is welcomed in the processing of digital data compression. The compression algorithm based on wavelet transform is also widely used in international image compression standards and the development of video data.As the theory "Context reduce the entropy ", people pay attention to context modeling based on wavelet transform and conditional entropy. However, when the Context modeling increases, the conditional probabilities can not be sufficiently counted, that's called "Context dilution ".To solve this problem,the majority of researchers introduced the Context quantization. Context quantization is thought to reduce the number of context probability distribution in order to avoid dilution. There are some classic Context quantization algorithm:Minimum Conditional Entropy Context Quantization algorithms, Maximum Mutual Information Context quantization algorithm, ant Colony algorithm. All the algorithms above proposed effective method to calculate the similarity between Context probability distributions, and clustering. But there is a common problem among the algorithms-parameter adjustment, people need to give the levels before quantization.The Minimum Description Length Context Quantization algorithm is described in detail in this paper. This paper proposed to use the MDLCQ algorithm to quantify the Context modeling.MDLCQ algorithm overcame the common limitation, it can find the best quantization level adaptively. The MDLCQ algorithm is applied to image compression based on wavelet transform in this paper, the experimental results show that the MDLCQ algorithm does not depend on human experience can achieve a good performance compared with other algorithms.Images show different amplitude characterristics in different sub-band area. Even a very eficient Context mideling cant not take all different frequency sources in account. Traditional encoders use only one Context modeling to compress iamges, but this paper proposed a multi-modeling hybrid coding scheme.So that the source could automatically select its best mode based on the probability statistics. By the experiment of image compression, we found that multi-modeling encoder is more efficient than a single one.