Image Denoising And Restoration

With the rapid development of science and technology, digital image processing has been widely applied to various fields of people life. The image is interfered by all kinds of noise in the process of storage and transmission, which leads to a low quality image. To improve the accuracy of image recognition effectively, image denoising is one of the main tasks of image processing. It provides reliable guarantee for subsequent image processing. This paper focuses on the study of the salt and pepper noise and the gauss noise.For the salt and pepper noise, an adaptive denoising algorithm based on pulse coupled neural network is proposed. The algorithm selects the size of neighborhood based on sub-image feature. Noise pixels are detected and smoothed through pulse coupled neural network in the neighborhood. For the image with more details, the size of neighborhood is determined by noise intensity. For the image with fewer details, the size of neighborhood is determined by texture complexity. Compared to median filter and traditional PCNN, the method performed better in removing noise and preserving image details. In addition, the algorithm was applied to the actual image with noise and obtained a good recovery image, which verified the practicability of the algorithm.For the gauss noise, this paper proposes a new denoising algorithm based on adaptive dictionary and sparse representation. The algorithm obtains sparse representation of images by training sub-dictionary for all similar sub-blocks. Firstly, the signal sub-space is determined through global analysis. Secondly, the image is divided into smooth area and texture details through discrete cosine transform. Then dictionaries of texture details and smooth region are trained through K-singular value. Finally, the high quality image is reconstructed using sparse decomposition. Compared to the method using fixed dictionary or global dictionary to recover the image, the algorithm had a good ability to preserve image details. The peak signal to noise ratio image had a significant improvement which verified the effectiveness of the algorithm.