Image Coding Algorithm Research Based On Multiwavelets And Neural Networks

In order to save, processes and transmits the large capacity image information, must compressed to the image information, also caused the image compression into research the importance domain.The neural network parallel processing ability, the high relevancy and the auto-adapted ability have decided its suitable image data compression. The auto-adapted PCA neural network has solved the KL transformation existence problem well, and could adapted the image change, moreover does not need anew train the network weights value. The SOFM algorithm compares the LBG algorithm this to form the aspect in the vector quantization code to have many merits, the initial condition sensitivity is low, can have the codebook which low distorts equally this, obtains the widespread application in the image vector coding.The multiwavelets take the wavelet analysis the development, not only maintained the wavelet merit, moreover has overcome wavelet some flaws, at the same time has had when the processing image requests the orthogonality, tight compactness, the symmetry, high vanish rectangular. The multiwavelets in the superiority which theoretically displays as well as it the potential which (image compression aspect) has in the application domain, this had decided it more and more is obtained extensive research and the application.The first part work of this paper is discussed the Principle Components Analysis (PCA) and Self Organizing Feature Mapping (SOFM) the network characteristic, proposed classified PCA/ SOFM mix neural network image vector quantization method, and compared with other neural network code methods, and at last analyzed and concluded.The second part work of this paper analyzes applications of multiwavelets on image compression code , proposed based on the energy and the maximum coefficient distribution image compression algorithm, and compared analysis based on the energy distribution method;The energy concentrated after the multiwavelets transformation in the low frequency sub- belt and destroyed between the SPIHT code request different sub-belt the element spatial relevant characteristic, proposed the improvement SPIHT code method, and compared with the standard SPIHT code method, and analyzed and concluded;After balanced multiwavelets transformation the different direction sub- belt coefficient relations, vector quantization of SOFM according to three directions,has realized the parallel processing, applies to in the different image code, and analyzed and the concluded.The experiment that used different image and different multiwavelets transformation image compression method is proofed that the neural network and the multwavelets has the very big superiority and the potential on application of image compression code.