Research On Multi-frame Image Restoration

As one of the significant media which carry the visual information for human-beings, digital images are tend to be influenced by noises and blurring during the acquisition or transmission, which results in a decline in the quality of images and prevents people from obtaining the visual information they need. How to effectively suppress the influence of noises and blurring on the image quality, and to provide more realistic and reliable image information for subsequent image processing, has always been considered as one of the key areas and hot issues in the research field of digital image processing. The traditional image restoration methods usually focus on one kind of degenerative phenomenon, such as noise or blurring. But in practice, noise and blurring are always accompanied by each other, which means that the presence of noise would impact on the elimination of the blurring, vice versa. In mathematics, due to the ill-posed nature of the classical single image restoration problem, one can hardly get the ideal image. However, compared to the single image, multi-image of the same scene captured in a short term can offer much more spatial-time redundancy information, which can be used for image restoration in a different way.In this work, the problem of image restoration by using multi-image of the same scene is addressed. Based on the spatial-time redundancy information provided by multi-image of the same scene, a method for jointly image de-noising and de-blurring is proposed. Since the noise has a strong influence on calculating the correlation between images, which is a key factor in estimating the blurring kernel in the multi-image de-blurring method, the accuracy of the correlation between images is improved by suppressing the noise effect and the blurring effect iteratively in the proposed jointly image de-noising and de-blurring method. In the actual scene, because of the hand-shake and other factors, there exists a relative movement between multi-image of the same scene, which means that the proposed jointly image de-noising and de-blurring method could not be used in the actual scene directly. To address this issue, with analyzing the traditional image registration methods, we propose a block matching based image registration method, which can be cascaded with the jointly multi-image restoration method proposed to suppress the effect of both the noise and the blurring.Extensive simulation results show that compared to the existing multi-image de-blurring methods, the proposed jointly multi-image restoration method preserves the detailed structures of the original image efficiently, with significant improvements in the subjective image quality and about 4d B in the increase of PSNR. Moreover, by cascading with the block matching based image registration method, the proposed jointly multi-image restoration method based on block matching can solve the multi-image restoration problem with relative motion accurately and quickly, which makes the method with much more practical value.