Context-modelin Based Application In The Loseless Image Compression Of Entropy Coding

With the rapid development of network applications recently and multimedia technology, the transmission and storage of multi-dimensional image data has become a relatively hot research area. Because of the requirements of a number of important areas, image compression must be reversible, that is enables lossless compression. So it owns important scientific meaning to research image lossless compression. As Context-based entropy coders can achieve high compression performance, it has had an important role in the coding field. With the increasing complexity of the informant source, Context-based source coding theory is increasingly widely used.Context-based image compression coding is mainly use the theory "conditional entropy is not larger than image entropy"in information theory in the experience of the sampling point in front of the sampling point.Self-adaptive Context modeling can effectively reduce the code length, that effectively reduce the entropy. However, it brings a inevitable question:modeling dilution. When the Context template is too large to provide enough sampling points by information source, the coding length will increase and thus reduce the compression perforation. So how to build Context modeling is the most important research direction of my thesis. In order to reduce the cost of the modeling and use more known conditional information, my thesis will bring the former learning experience into the latter pixel coding based on the method of training set based on Context-modeling.The Context-modeling in my thesis is will match to pixels similar to current sampling point as training set through Context modeling. Then I will self-related rank the Context modeling points of the pixels in the training set. And in the end I will determine the order of the model through the self-adaptation of the minimum description length. In this way, pixels with slowly image change will choose a larger order, pixels with drastic image change will choose a smaller order on the contrary. Taking advantages of prior learning experience, the modeling size in this thesis is determined by current statistics, solving the modeling dilution caused by high-order entropy. I also encountered a problem that the training set data of the pixel is too little in coding, resulting in low coding efficiency. My thesis will initialize probability distribution with the combination of GGD algorithm and Context modeling, further optimizing the statistical probability distribution. The experiment proves that the entropy coding algorithm based on Context modeling do have the validity.