Research On Restoration Method Of Motion Blurred Forestry Images

With the development of modern forestry,intelligent forestry equipment has become an inevitable trend.For forestry robots,machine vision system has become an important way to obtain the information of working environment and realize intelligence.However,due to the complex and changeable working environment,rugged roads and ubiquitous obstacles,it is difficult to avoid the bumps and vibrations of the forestry robot.That leads forestry images taken by its vision system blurred,thus seriously affects the subsequent recognition,reconstruction and other tasks.Therefore,it is necessary to research on restoration method of forestry motion blurred images.At present,most of the blurred image restoration methods are aimed at general natural images or blurred images in other specific fields.The research on restoration of forestry blurred images is rarely involved.Based on the particularity of forestry images,this paper studies the restoration of forestry motion blurred images.The main research contents and innovations are as follows:1.Aiming at the abundant edge texture in the forestry image,a method of blurred image restoration based on edge selection and sparse representation is proposed.By fitting the gradient distribution of forestry images,the algorithm combines the Hyper-Laplacian prior of image gradient distribution with the block-wise sparse representation.The global gradient constraint enhances the smoothness between adjacent pixels,and effectively reduces the disadvantage that sparse representation is easy to produce artifacts at the edge of the block.In addition,the algorithm proposes an effective edge selection method,and introduces an intermediate image to iterate,which reduces the interference of various complex textures in the forestry image to the estimation of blurring kernel,and thus improves the accuracy of the estimated kernel.The algorithm realized effective restoration of the forestry gray-scale blurred image.2.Considering the abundant different scaled edge textures of forestry images,the blurring kernel is solved by adopting a scale-invariant normalized norm prior and blurring kernel constraints.Then,by combining the Hyper-Laplacian prior of image gradient distribution with the block-wise sparse representation and using it as a non-blind restoration algorithm,a regularization restoration method of forestry motion blurred image is proposed,which has achieved good restoration results on color forestry blurred images.3.Forestry images are easy to be affected by the environment lighting,but the color information of an image maintain constancy.Thus,a motion blurred image deblurring method using edge-based color patches is proposed in RGB color space.By analyzing the distribution of color pixels in the image blocks at the significant edge of the image,and combining with the two-color model,a weighted colorregularization prior term is proposed.Based on this prior,a blurring kernel estimation model is constructed.Due to the use of color image blocks,the color information in the image and the relationship between adjacent pixels can be made full use of,so that the estimation of blurring kernel is more accurate and stable.4.According to the large number of repeated structures of same scale and cross scale in forestry images,an image deblurring method based on cross-scale non-local similar patches is proposed.In this algorithm,a prior term of low rank matrix constraint based on cross-scale non-local similar patches(CS-NSP)is constructed by using the pixel values based on a prior feature that the down-scaled blurred image is clearer than the original blurred image.By combining the CS-NSP constraint of image gradients and the sparsity constraint of gradients respectively,CS-NSP*2 algorithm and a faster CS-NSP*2-L0 algorithm are proposed.The deblurred results on the test dataset showed that the proposed algorithm can effectively restore the blurred forestry image,and the deblurring result is more advantageous at the part with fine structures.