Adaptive Neighborhood Inpainting Method For Depth Image

With the progress of optical technology,the method of gathering depth information has become more and more mature.It also plays an increasingly important role in the application of computer vision,such as three-dimensional reconstruction,human-computer interaction,target tracking,free-view video rendering and reality enhancement.However,Kinect and other consumable depth cameras which gather depth information by active sensing technology have a lot of noise in their depth images,which seriously affects the use of depth information in practical applications.In the intelligent development of workshop,a lot of depth information is needed,such as dimension detection of parts,object grabbing of mechanical arm,etc.Therefore,it is necessary to inpaint the depth image to obtain more accurate depth information.In this paper,we mainly improve the effect of depth image inpainting in two directions:1)The order of image inpainting.The order of image restoration almost determines the result of the texture propagation,and more accurate repair order often means more accurate repair results.Based on the Criminisi algorithm,this paper defines a new method of calculating confidence term by using Euclidean distance,and improves the priority to weigh confidence term and data term by analyzing the changing trend of confidence term and data term through the iteration process.Priority is introduced into the neighborhood inpaint algorithm for depth image restoration,and it is found that the texture extension of the inpainting results guided by priority is more reasonable.2)The adaptive neighborhood of the pixels to be inpainted.Traditional neighborhood partitioning contains pixels whose depth information does not match the pixels to be inpainted,which cause the inaccurate pixel estimation.The proposed adaptive neighborhood of the pixels to be inpainted is to find the pixels which more similar to the pixels to be inpainted and use them to estimate.The criterion of region growth is perfectly suitable to search two pixels with high similarity.By the guidance of the color image which registered with the depth image,we find a set of pixels with high similarity to the pixels to be inpainted,i.e.adaptive neighborhood,to replace the traditional neighborhood,and inpaint the depth image under the control of priority.Compared with other inpainting algorithms,the experimental results show that the adaptive neighborhood can ensure more accurate inpainting results.