Research On The Application Of The Stochastic Learning Weak Estimatorson The Probability Estimation In The Adaptive Entropy Coders

Entropy coding, in particular adaptive entropy coding, is a core part of most image and video compression coding standards’ and many non-standards’ encoder, such as the MQ coder in JPEG2000 and the CABAC in H.264 / AVC,both encoders use adaptive binary arithmetic coding. The performance of entropy coding is mainly related to two factors, one is the probability model consistent with the actual characteristics of whether the source of the second encoder to be based on a probability model for the way the code symbol assigned codeword. For the second point, at present, when a given probability model, a lot of average code length coding method is already very close to the Shannon entropy given probability model, the performance in this regard has been very limited promotion. As for the first point, when the statistical characteristics of data to be encoded more stable, establish a probability model problem is relatively easy, but when the data to be encoded statistical properties change frequently, probability models tend to correspond to the actual characteristics of the source of some deviation, affect the encoding performance, if the probability model can reflect real-time changes in the encoded data to be such a feature, in theory, can achieve better coding results. Against this background, the establishment of the time of this question often changes the characteristics of the data probability model entropy encoder launched a study.Firstly introduces the definition of entropy coding are closely related to the source of entropy, probability model and estimate the entropy, analyzed the relationship between entropy source entropy coding code length and data to be encoded and the estimated probability model entropy constraint between explain the adaptive entropy encoder probability estimation model an important role. Based on the above basic theory, then give the definition of the data in this article smooth and non-stationary data and related mathematical representation and analysis of the static model and Bayesian parameter estimation based on the traditional theory of probability estimation algorithm is not suitable for non-stationary data probability estimates reasons and relevant experimental verification, and finally introduces the probability windowed law and forgetting two classical factor analysis of non-stationary environment estimation algorithm, and analyzes the probability for each probability estimation algorithm estimates the effect of the different characteristics of the data, Finally, we discuss the effect of non-stationary process data characteristic change of intensity of an estimate of the overall probability, put forward the basic idea of change characteristic data is adjusted according to the probability estimation algorithm adaptive capacity.Secondly, the study of weak random estimation theory(SLWE) in the non-stationary data probability estimation application. First introduced on SLWE binomial and multinomial distribution parameter estimation process and the qualitative and quantitative analysis of both its applicability, in-depth analysis of the principles inherent probability updated, and compares it with the windowing method, analysis its internal relations with windowed law. Finally Region Coding architecture using SLWE interval probability estimation algorithm coding method, described in detail the range of problems may be caused by degeneration when transplanted into the interval SLWE algorithm coding and decoding due to the floating-point rounding error caused by the accumulation of problems and corresponding solutions. Finally, the experimental analysis of the effects of the new coding interval encoding method different characteristic data.Finally, the adaptive algorithm parameters adaptability SLWE adaptive algorithm based on the local characteristics of the data changes. The algorithm for the actual data change characteristic changes in the characteristics of more complex, the first use of the local statistical characteristics of the data analysis of the data of non-stationary, then extract the characteristic change of location as a relatively large change point, according to the position change last point and the degree of characteristic change of the adaptive learning algorithm SLWE factor corresponding change algorithms adaptability and convergence ability to adapt to local characteristics of the data changes. Finally, the experimental analysis...