| This paper mainly proposed the image compression methods in which we combine the integer wavelet transform with the vector quantization under the lifting frame.These methods are as follows. First, the image coefficient is properly amplified (this article amplified by 1600%), in order to reduce mantissa cutoff `s influence on the transform coefficient amplitude. It uses Db9/7 wavelet filter to achieve three level integer wavelet transform. Second, it performs the DPCM coding on the low-frequency wavelet coefficients, whereas the high-frequency wavelet coefficients perform the address rearrangement, so as to get the tree structural vector. The quantization part adopts the local search method, adjust the restoring image quality by adjusting the range, in order to increase the coding speed and restore the image quality.In modern communication, image transmission is the important content. The size of transmission information content is one of the most important reasons influencing the speed. In order to heighten the communication velocity, one necessary means is that it adopts coding compression technique and reduces the transmissing data quantity. The compression of image data has already an urgent need for technique progress. For this need, its algorithm and technique has already been an extremely active research field in the recent 30 years, also achieves great success in commerce. We have many coding methods in the traditional and developing respect. The traditional methods, for instance, are PCM quantization, spatial and temporal subsampling coding, entropy coding, predictive coding, transform coding, vector quantization(VQ), and subband coding. The developing ones are fractal coding, model –based coding and wavelet coding. Various methods have their distinctive parts, also have defects. If we combine the traditional methods with the developing ones, the image compression quality will be better.The spatial distributed characteristics and high resolution ratio of the wavelet image coefficients lead to correlation in spatial position and content,so adopting VQ technique is suitable. Not only it eliminates redundant code and complexity from the loose spatial distributed structure, but describes the correlation among the different coefficients. Moreover, combining wavelet transform with VQ technique can avoid three problems in the VQ coding. Firstly, it makes it difficult to produce the general code. Secondly, it reduces the resolution ratio for the LBG algorithm `s smoothing on the high frequency component. Thirdly, it cannot link with the human `s visual system characteristic. The scheme for the image compression coding on this paper is that we combine the integer wavelet transform with the VQ image compression under the lifting scheme. IWT achieves great success in the lossless coding and compression, but its efficiency in the loss compression is far lower than the traditional DWT. Especially based on IMT for the rational transforming parameter, PSNR of the destructive compression differ 3~6dB from DWT adopting Db9/7 under the SPIHT and SPECK coding frame. There are two reasons. First, that mantissa rounding makes IWT become the nonlinear transform, thus reduce its energy centralization. Second, based on the above problems, above all this paper properly amplifies the image coefficients of the _ so as to reduce mantissa rounding`s influence on the transforming coefficients` amplitude .Afterwards, introduces panto factor according to the rational integer IWT in order to raise IWT `s energy centralization. Finally, through adjusting panto factor, makes IWT `s destructive compression achieve the optimum results. Choose a normal lena of 512 pixel multiplied by 512 pixel, the Lena `s pixel is amplified by 1600%. Then perform three level integer wavelet transform and choose Db9/7 wavelet filter. Choose ten representative static image as the training sequence, produce the vector code using LBG algorithm, order the code according to the root node size of the vector, form the ordered code. The quantiza...
|
PDF Full Text Request