Investigation Of Image Denoising Based On Structure Prior And Sparse Representation

As a fundamental topic in the field of computer vision,image denoising has always been a hot topic in academic research.Image noise generally affect the accuracy of image postprocessing such as image segmentation,image encoding,feature extraction,and target detection.While the image contains both of rich texture information and geometric information,most of the existing image denoising algorithms can not simultenously preserve the image information as well as remove the image noise effectively.In response to this problem,this paper introduces the structure prior and sparse prior of image into the image restoration process,and then proposes two improved algorithms based on total variation and sparse representation.Aiming at the problem of noise residual and over soomth effect existed in infrared image denoising algorithm,this paper proposes a non-uniformity noise removal method based on sparse representation.The method is based on the sparse prior of non-uniform noise.The learning dictionary is trained by the K-SVD algorithm,and the learning dictionary is sorted according to the strip noise characteristics.The infrared image is sparsely represented using the rearranged learning dictionary,and the noise is removed on the sparse coefficient of the image to remove.Finally,for the problem of large noise residual and image blurring,the non-uniform noise removal of the mean value of the image block is also carried out,which improves the ability of the method to remove non-uniform noise.Aiming at the problem that the existing Gaussian noise removal algorithm has insufficient ability to maintain details,this paper proposes a total variation image denoising method based on non-local steering kernel.The method utilizes an image self-similarity prior and the guiding kernel with excellent texture feature retention.By combining non-local weights with the guiding kernel,a non-local guiding kernel is proposed.This non-local guiding kernel is more accurate in describing the image structure and is more resistant to noise interference.Especially in the texture area,the weight is consistent with the height of the texture feature.The non-local guidance kernel is used to construct the weight of the total variation regularization term,which effectively improves the ability of the total variation method to remove noise and maintain image details.The results of simulation image experiment and real image experiment show that the nonuniformity noise removal method based on sparse representation and the total variation image denoising method based on non-local steering kernel has good ability to remove noise,effectively improves the peak signal-to-noise ratio and structural similarity of the image,and maintains the details of the image edge and texture in the visual aspect,resulting in a sharp visual effect.The research results in this paper also have important guiding effects on restoration applications such as image deblurring and super-resolution reconstruction.