Moving Object Segmentation Based On The Model Of 3D Visual Attention And Its Application

Visual Attention is one of the most important mechanisms of Human Visual System, which helps human eyes quickly focus on the most salient regions of complex scenes. Visual Attention Model is established by simulating the Visual Attention Mechanism, and the corresponding mathematical model is established, which can effectively detect the object or region of interest in complex visual scenes, it shows great advantages and application prospect in image retrieval, object detection, image segmentation and many other research regions, and recently Visual Attention Model have attracted lots of attentions from researchers all over the world.Recently, a lot of Visual Attention Model for 2D images has been proposed, however,2D visual attention model showed great difficulty when dealing with 3D images and videos, because the lack of research in depth information. By concerning the shortcomings of existing model, the Visual Attention Model combined with depth information is studied in this thesis, and an effective saliency detection method for 3D images and videos is proposed. And finally, the application of 3D visual attention model in moving object segmentation is also studied. The main contents are as follow:(1)The basic theory and methods of visual attention model are studied. The basic mechanisms of human visual attention are discussed, the common 2D visual attention models and saliency detection methods are summarized, and finally the evaluation methods of saliency detection is studied;(2)A 3D visual attention model combined with depth information is studied. The extraction method of depth information and its influence on human visual system are studied, the common 3D visual attention models and their construction methods is discussed, and a 3D visual attention model based on image block segmentation is studied finally;(3)A 3D saliency detection model based on superpixel segmentation is proposed. An improved superpixel segmentation method combined with depth information is proposed based on the study of superpixel segmentation algorithm for 2D images. And a 3D saliency detection method based on the improved superpixel segmentation is proposed. And finally, a refining method for saliency detection results based on superpixel graph model is proposed.(4)The application of 3D visual attention model in moving object segmentation is studied. The common moving object segmentation methods and the motion information extraction methods is studied. A moving object segmentation method based on the motion information assisted 3D visual attention model and Growcut algorithm is proposed;...