Research On Lossless Coding For Video Compression

Lossless coding is a process which maps the original source into code words without any information loss.The most widely used lossless coding method in video compression is Context Adaptive Binary Arithmetic Coding.In this method,context modeling and binarization are firstly used to approximate the original source as a stationary stochastic process.Then arithmetic coding uses the binary conditional probability distribution in the stationary stochastic process to map original source into code words.This paper studies the key problems in lossless coding,such as source modeling,context modeling,binarization and arithmetic coding.Log-domain binary arithmetic coder and decoder avoid the use of multiplier and are simple for implementation.However precision of their variables grows without bound for long sequences.This paper proposes a fixed-precision log-domain binary arithmetic coder which has three improvements:interval scaling,bit-stream early output and outstanding bits reset.This paper also proposes a fixed-precision log-domain binary aritlhmetic decoder which has three improvements:interval scaling,bit-stream batch reading and log-domain interval comparison.Besides,an encoder side constraint scheme,which constrains the number of successive Most Probable Symbols(MPS),is also proposed to constrain precision of decoder variables without any decoder changes.The encoder side constraint scheme is adopted into software of AVS1/IEEE1857.1 video compression standard.Ignoring the influence caused by probability estimation,context modeling can increase coding efficiency.Considering the storage overhead of context modeling,this paper proposes a design criterion for context modeling which aims at low storage overhead and high coding efficiency.Based on the same assumption,binarization cannot change the coding efficiency.Consider its influence on throughput rate of lossless coding,this paper proposes a design criterion for binarization which aims at high throughput rate.Based on the proposed design criteria,this paper designs context modeling and binarization algorithms for syntaxes in video compression.The storage overhead is reduced and throughput rate is increased without any coding loss.All the design results are adopted into AVS2/IEEE1857.4 video compression standard.Previous lossless coding theories model the original source as a stationary source.This paper proposes to model the original source as a non-stationary source which is in line with actual situation.Probability distribution,which is a large-scale property,is not suitable for describing non-stationary source.This paper proposes to use possibility distribution instead of probability distribution to describe non-stationary source.Arithmetic coding uses estimated possibility distribution to generate code word.The correlation between neighboring samples is exploited to compress the source.Code rate is the cost for average possibility distribution prediction error.This paper also proposes two methods to reduce average possibility distribution prediction error.Using the proposed methods in video compression system,prediction error is reduced and coding performance is significantly improved.