Researches On The Application Of Gradient Features In Image Denoising

The gradient features of the image play an important role in image denoising,which reflect image edge,contour and texture information.In this paper,the applications of image gradient features are studied in image denoising,and effective image denoising methods with gradient constraints are proposed,which can preserve edge,contour and texture of image and enhance image clarity.The specific work of this paper includes:Firstly,a multi-direction gradient constrained total variation denoising method is proposed.As the traditional denoising methods only exploiting horizontal and vertical direction gradients are not perfect,so gradient features can be extracted by gradient operators from horizontal,vertical and two diagonal direction respectively.Then,the gradient features from multi-direction are used for total variation denoising method as constraints persuing better results.Experimental results show that the improved method with gradient feature constraints is better than traditional method constrained by only two direction gradients and can obtain higher PSNR value.Secondly,a subregional image denoising algorithm based on multi direction gradient features is designed.Image is segmented into two regions due to different gradient features,which are smooth region and edge region.Smooth denoising methods are used to remove noise in smooth region and denoising methods preserving edge and texture are used in edge region.Experimental results show that,compared to the overall denoising method,the subregional denoising method can suppress the noise effectively while preserve the details of the edge region and improve the overall denoising results.Thirdly,subregional denoising method based on gradients constrained total variation is designed.The gradient distribution characteristics of the noise free image and the noisy image are analyzed.After image segmentation,gradients of different regions are shrinked by different gradients shrinkage method.Gradients of smooth regions are shrinked linearly while gradients of edge regions are firstly shrinked linearly and then sharpened.Most of the noise in image can be removed by overall gradient shrinkage,and gradients of edge regions are sharpened to eliminate blur effect by overall shrinkage.The experimental results show that the subregional denoising methods can obtain higher PSNR.