The Blurred Image Reconstruction Based On Camera

Images are often degraded by imaging devices and external environmental noise when they are being collected, transformed and transmitted. For example, when we use digital camera shooting, the assemblage of optical systems, the deviation of shooting and focusing, the camera shake at the time of exposure or target moving etc. all make the image blurred. This paper focuses on the research of removing the image blurring caused by the camera shake. In most basic theories, there is no definite solution to the problem because we know the original image and the blur kernel very little, all the practical solving schemes must make very strong prior hypothesis; when solving the problem, conventional signal processing usually make a universal hypothesis in the form of frequency energy law. The relevant algorithms can only process very small size blur kernel, but they're helpless to the complicated one associated with camera shake.Based on the method of restoring the image by estimating the camera shake trace which is proposed by Fergus in 2006 as reference, we find the difference of statistical characteristics between blurred images and latent images by exploiting the natural image statistical characteristics more deeply, i.e. the precise components of natural clear images obey the Laplace distribution in their wavelet domain, but blurred images don't have this kind of characteristic. We take the specific distribution of precise components in wavelet domain as a prior information, estimate the unknowns by the Bayesian approach to find the blur kernel, and then reconstruct images using a standard deconvolution. The experimental results can be judged in three aspects: convergence, image definition and texture. The experimental data shows that the quality of the restored images by our algorithm is obviously better than the one of original images. Additionally, on the similar restored results, the convergence of our algorithm is much faster than Fergus's algorithm.Most imaging devices introduce some form of luminance and geometric nonlinearities. Therefore we also need to consider the influence of the two kinds of nonlinearities to image restoration. Since people are more sensitive to luminance distortion compared with geometric distortion, we only consider the influence of the luminance nonlinearities to image restoration in this paper. We mainly research the blind removing of luminance nonlinearities. By analyzing the specific high order (more than two orders) correlation of image spectrum caused by nonlinearities, these correlations can be estimated using tools from polyspectral analysis. The nonlinearities can then be estimated and removed by simply minimizing these correlations. The experimental results show that our algorithm improves the restored results.