Adaptive Arithmetic Coding Method Based On Neural Network And Maximum Entropy Principle

Universal lossless data coding is a branch of the data compression research. Traditional coding algorithms are based on simple n-gram models. Those distant constraints and some redundancies in natural language, such as word order redundancies, semantic redundancies and syntactic redundancies, are neglected in the n-gram models. In order to find and reduce those redundancies, we need more intelligent algorithms based on more effective models. The Maximum Entropy Principal proved to be a very useful method to create statistical language model. Neural network, which is used widely in many research fields, can do self-adapting and self-learning, so they become a preferable approach for data coding.Neural networks are used more frequently in lossy data coding than in general lossless data coding, because standard neural networks must be trained off-line and they are too slow to be practical. In this thesis, statistical language model based on maximum entropy and neural networks are discussed particularly. Then, an arithmetic coding algorithm based on maximum entropy and neural networks are proposed in this thesis. This adaptive algorithm with simply structure can do on-line learning and needn't to be trained off-line. The experiments show that this algorithm surpasses those traditional coding method in compressing rate and it is competitive in speed and time with those traditional coding method.