The Research On Color Image Restoration Method Based On Harmonic Neural Network

In recent years, the rapid development of computer technology has brought a wave of research of image restoration technique.Image restoration technique has widely applied in medical imaging, astronomical imaging, military, image and video coding and other areas. However, the quality of image will be reduced inevitably during the process of imaging, copying and translating of image. Therefore in the many areas, it is necessary that the image information is clear and with high quality. So that image restoration is very meaningful. Image restoration problem is an aspect of signal processing field, it can be seen as a deconvolution problem of signal. Because of the nondeterminacy of image quality reducing type and the pollution of noise, image restoration problem is a type of deconvolution problem with ill-posedness.Based on analyzing the basic theory of image restoration technology and neural network in depth, and analyzing the essential principle and the merit and demerit of several classic image restoration method of tradition and modernization, the paper gives out a new image restoration method of harmonic model neural network combined with color image restoration problem:The thesis analyzes the subjective and objective evaluation method at first, a new comprehensive assessment method of image restoration quality based on structural similarity is proposed on this basis. According to the visual feature of human eyes:the comprehend of visual perception and figure mainly depends on organic structures. So that evaluation method divide the quality reduced figure into fringe area, flat area, texture area, and then calculate the correlation coefficient of the original image and the restored image of the areas respectively, then adopt different weighted values and calculate the similarity of the original image and the restored image according to the importance of different areas, gives out the calculating method of the comprehensive assessment the quality of image restoration.Then improve the quality and efficiency of restoration method from two aspects:one is degraded image, the other is restore tool. Now that image restoration problem is a kind of deconvolution problem, then it is necessary that improve the quality of original input signal and the calculating speed. Therefore, the thesis reconstruct the degraded image at first to increase the SNR of the original input signal, and gives out a color image reconstruction method based on the RGB of single frame degraded image. Then the merit and demerit of existing regularization image reconstruction method is analyzed, and the regularizing operator is improved, and combined with neural network, the algorithm not only improves the convergence speed, but also improves the property of fringe restoration and fake image compared with other regularization methods.Finally, the research work of this task is summarized, and the research direction in the future is proposed.