One Algorithm Of Image Compression Based On Wavelet Transform And Vector Quantization

With the rapid development of computer multimedia, information andnetwork technologies, there have been more and more demands for usingdigital devices to store, process and transmit images in digital style.Because of the large amount of digital images, huge storage space and widebandwidth are required. The study of the image coding algorithms and itsapplications is one of the most active areas in the information technology.Due to multi resolution analysis, the image compression algorithm based onthe wavelet decomposition is an efficient method with high compressionratio and small reconstruction difference. On the other hand, vectorquantization (VQ) is an efficient compression technique, whose prominentvirtues are high compression ratio and simple decoding process, so it hasbecome one of important compression techniques. VQ has been used forimage compression and speech coding successfully and has bright prospecton the compression and real-time transmission of satellite-sensed, imagedatabase and so on. So, it is important and imperative to develop the studyof image compression algorithm and its applications based on wavelettransform and vector quantization for higher compression ratio, good imagequality and smaller time complexity.According to the advantages of high efficiency of wavelet transformand VQ in the field of image compression, this thesis researches the imagecompression algorithm based on both of them in order to improve thecompression ratio and reduce the time complexity of VQ.Before scalar quantization to the low frequency sub-band of waveletimage, normalization was performed firstly. Compared with the DPCM, thismethod is convenient to realize and has higher compression ratio with littledecrease of image quality. The disadvantage of distortion accumulation canalso be prevented. Vector quantization is used to compress the high frequency data. Inorder to reduce the time complexity of VQ, this thesis focuses two aspects.A new method of vector classification is presented to improve the efficiencyof forming the original codebook through reduce the vector number in thetraining sets. This method takes the advantage of direction and correlationbetween wavelet coefficients. On the other hand, optimal pair nearestneighbor algorithm is presented to reduce the time complexity of VQ.Test results show that the proposed algorithm based on wavelettransform and vector quantization improves not only compression ratio butalso efficiency through vector classification and optimal pair nearestneighbor algorithm.