Research On Denoising,Sharpening And Color Enhancement About Digital Image Enhancement

With the rapid development of digital imaging technology, digital images appear in every field of our daily lives. However, because of the equipment limited, environmental disturbance, and human factors, the acquired images are often influenced by noise, fuzzy and color problems. The existence of these problems not only make us cannot observe clearly natural images, but also bring many difficulties to the subsequent image analysis and recognition. Thus, this paper has carried on the research and discussion on the three aspects, image denoising, image sharpening, and color enhanced, and the main achievement is as follows:(1) Based on the sparse and redundant representation theory, this paper proposed improved adaptive backtracking-based orthogonal matching pursuit (IABOMP) algorithm. First of all, the adaptive threshold determined by the dictionary atoms together with the image content is applied, which makes the algorithm more intelligent. Secondly, the support set verification and residual feedback mechanism make sure the reconstructed image preserve as many as possible image details and the representation coefficient most sparse. What is more, the temporary dictionary and adaptive step size techniques reduced half of the number of iterations during algorithm processing. The experimental results show that the proposed algorithm has obtained good performances in terms of signal reconstruction accuracy, coefficient sparse level, processing time, PSNR value, and method noise assessment.(2) For dealing with the images with strong noise and focal fuzzy, this paper proposed local structure adaptive sharpening (LSAS) algorithm. Compared to other methods, the LSAS can sharpen the image details along the edge. As a result, it can effectively avoid the problems of noise amplification and details blurring. Besides, the over-shot problem will not appear. In order to solve the problem of the sharpening effect is not ideal due to setting parameters only by experience, we have also expanded out two iteration versions based on the proposed LSAS for image blind sharpening. One of them gets result by changing the parameters of local image covariance matrix, named ILSAS. The other optimizes the smooth parameter directly for the final restoration, named OILSAS. The experimental results show that the two iterative versions can got the satisfactory results within a few iterations.(3) In order to avoid amplifying noise in dark areas while enhancing image details, this paper proposed two algorithms based on image intensity for high dynamic range compression. One of them is the adaptive value-based multi-scale Retinex (AVBMSR), and the other is the adaptive multi-scale tone mapping (AMSTM). Different from the average weight generally used by other methods, the adaptive weight assigns unique value for each pixel of each scale output. Besides, the proposed algorithms can successfully avoid the problem of hue shift and color desaturation without the color restoration processing. The experimental results show that the proposed algorithms has obtained good balance in terms of high dynamic range compression, local details enhancement, color rendering, and noise suppression in dark areas, which makes the results have better visual effect.