Image Segmentation And Its Applications In Image Depth Estimation

Image depth estimation is a kind of technique to estimate scene depth from image,which is a form of image understanding.When the depth is extracted from the scene,the depth estimation results often suffer from problems such as fuzzy structure,indefinite level,object depth and background fusion.In order to solve the above problems,it is a feasible scheme to obtain different scene regions by image segmentation,and to carry out depth estimation and optimization.In this thesis,the image segmentation technology is studied and used in the depth estimation method,in order to improve the quality of the depth estimation.In this thesis,two kinds of segmentation methods are studied,which are used to deal with the segmentation of target objects and image scenes.Aiming at the problem of object depth disappearing,this thesis proposes a video moving object segmentation algorithm,which uses the objectness technique to extract the object area from the video and segments the object by analyzing the motion and color of the region.Combined with the results of object segmentation,this thesis uses the depth relation between the object and its support surface to optimize the depth estimation result,which solves the problem of the disappearance of the object depth.Aiming at the problem of fuzzy scene structure on the depth map,this thesis studies the image semantic segmentation method and realizes the semantic analysis of the image scene.The algorithm uses the convolutional network to classify all the pixels of the image,obtains the initial segmentation result,and then uses the overall modeling ability of the random condition random field to optimize the segmentation result.In this thesis,semantic segmentation is applied to the depth estimation,and a new depth estimation algorithm is proposed.The algorithm uses semantic segmentation to complete the scene analysis,and uses the structured random forest algorithm to establish the depth estimation model for different semantic regions respectively.Experiments show that the algorithm can get a clear structured depth of the scene,with high accuracy.The research and experiment of this thesis show that the image segmentation technique can help to improve the depth estimation quality,which can play an important role in solving the problem that the depth of the object disappears,the fuzzy scene structure and the poor depth level.