The Inpainting Of Low Precision Depth Image Based On The Minimum Spanning Tree Image Segmentation

Depth information has been widely used in target recognition,3d reconstruction,robot navigation,3DTV and other fields.With the release of low-cost depth sensors such as Microsoft Kinect,the field of depth information applications has undergone significant changes.These sensors can be used for real-time 3d reconstruction,skeleton tracking,image recognition and other fields due to their capacities to access real-time depth information and color information.The loss of information in depth data collected by these sensors cause much noise and large invalid regions in depth images,which limits its application.Therefore,it is necessary to inpaint these low-precision depth images for obtaining high-quality depth images.Firstly,this paper introduces the background of depth information and the significance of depth image inpainting,then explains the working principles of Kinect depth sensor,and analyses the causes of the void area in depth image and the difficulties of inpainting.Based on Kinect device,this paper studied the depth image inpainting algorithm.The main research and achievements are as follows:(1)A depth image inpainting algorithm guided by the minimum spanning tree image segmentation was proposed,and the image segmentation algorithm based on the minimum spanning tree was introduced to segment the color images,so as to construct the depth image inpainting guided by the guidance information;(2)The invalid regions are based on the inpainting sequence of the fast marching method,and the weight function of the fast marching method has been improved;(3)A depth image inpainting algorithm guided by local segmentation is proposed.By combining the idea of local segmentation,meaningful areas of color images for depth image restoration are extracted,which improves the efficiency of building guidance information.(4)A invalid regions' type discrimination algorithm is proposed to judge the types of invalid regions in depth images.In the end,we did some experiments through several depth images and color images and compares our algorithm with some popular depth image inpainting algorithms proposed in recent years.Through the comparison of objective quantitative indicators and subjective visual effect,the advantages and effectiveness of the two algorithms proposed in this paper are verified.