The Research Of Image Inpainting Algorithm Based On Hybrid Filtering

With continuous development of computer vision,depth images which are used to describe image depth information have received more and more attention.At the same time,the ways to acquire depth information have become increasingly convenient and accurate,and they have greatly promoted the research of stereo vision,three-dimensional reconstruction and so on.Due to the limitation of the equipment and the influence of the acquisition environment factors,there are some problems in the process of acquiring depth information,such as appearance of inevitable holes in the image due to lack of depth information,image noises,and random hops of depth values.In this thesis,according to the problems existing in the acquisition process of depth images,an image inpainting algorithm based on Hybrid Filtering are proposed to obtain more accurate depth information.Meanwhile,it also promotes the development of relative research for example stereo vision and three-dimensional reconstruction.The main work of this article is as follow:(1)This thesis proposes a depth image inpainting algorithm based on hybrid filtering to solve the hole problems in the depth map.Firstly,we propose to set priorities for all pixels in the hole for the overall hole-filling strategy.The priority depends on the support and credibility of the neighboring pixels for the center-cavity pixel.In the process of hole filling,the priority of the remaining pixels is constantly updated,ensuring that the pixels with the highest priority can always be filled firstly;(2)Based on the proposed priority of hole pixels,this thesis optimizes the calculation of depth values.In this thesis,based on the principle of joint bilateral filter,the total variation is used to extract the structure and texture of the image from the color image,and these two kinds of information are added to the calculation of the depth value weight,which makes the depth value calculation is more accurate and effective;(3)In the process of image inpainting,due to the difference of hole and noise and the particularity of edge pixels,all the hole pixels are divided into four categories according to whether they are in holes or whether they are located at the edges.The proposed algorithm is for improving the accuracy of edge pixels,reducing the filling error,enhancing the image repair performance;(4)In order to verify the effectiveness of the proposed algorithm,the algorithm proposed is compared with FMM and DJBF algorithms.Through the comparison of experimental results,inpainting algorithm based on the hybrid filtering not only improves the accuracy of the depth value,but expands the scope of the algorithm.