Research On Vector Quantization Technology For Image Compression

With the rapid development of the internet and the improvement of people's requirement of information,a large number of static images and dynamic images not only need to be store and process,but also need to be real-time transport.If there is no data compression is performed,the real-time transmission of the information can not be guaranteed under the existing communication conditions.At present,the storage and transmission of multimedia information have become a huge obstacle to digital communication.Therefore,the data compression has become more and more important.Vector quantization is an effective data compression technique with high compression ratio.It is not only used for image compression,but also widely used in the field of image retrieval,speech recognition and so on.There are three main vector quantization techniques: codebook design,codeword search and codeword index assignment.Because the codebook design is the basis of vector quantization,so it is the core of the entire technology research.This thesis mainly analyzes the existing codebook design algorithm and codeword search algorithm,and proposed a new optimization program.The classical LBG algorithm is simple and intuitive,and is widely used in codebook design.But because its initial codebook is obtained by the random selection method,the final access to the codebook performance is poor.In order to improve the quality of the final codebook,in this thesis,the particle swarm optimization(PSO)algorithm is applied to codebook design.Firstly,the initial particles are obtained by using the idea of consistency,so as to reduce the appearance of atypical primordial particles.Secondly,through the modification of the inertia weight and the learning factor,the early stage of the particles can be subject to their own information for global search,and the late can be subject to social information for local search.Finally,the updated codebook is obtained by the nearest neighbor condition and the centroid condition.The experimental results show that the improved algorithm is not only stable,but the quality of the codebook obtained by the improved algorithm is higher than the LBG algorithm.The characteristic quantities of vectors and the characteristic quantities of sub-vectors are related to the square euclidean distance of the vector,so they are widely applied to vector-quantized codeword search technology.In this thesis,we analyze thesearch range determined by the two feature quantities of mean and variance,and give a code word search algorithm based on mean,variance and trigonometric inequality.The experimental results show that the algorithm can effectively eliminate the codewords that can not be excluded from the elimination criteria determined by the two feature quantities of mean and variance,and the computational cost required for coding is low.Then,through the analysis and comparison of the search range determined by the sum of the vector and the value feature,the norm feature and the sum of the value of the sub-vector,this thesis presents a codeword search algorithm based on trigonometric inequality and sub-vector.The experimental results show that the algorithm is not only stable and the required computation is low.