Research And Implementation Of Image-based Rendering Technology

With the development of the society,the ordinary 2D display gradually cannot meet people's needs on the visual watching and 3D display is becoming a new trend.Multi-view 3D display is convenient to watch without other devices and the users can switch between multiple views,so it becomes a research hotspot and has important research significance.A key technology of multi-view 3D display is the virtual viewpoint generation.The technology synthesizes views from any desired perspective using a set of real-scene viewpoints.The existing virtual viewpoint generation is mainly Image-Based Rending(IBR)technology.The work of this paper includes research on a variety of IBR technology and study of the theoretical basis of technology,and especially analysis of Depth Image-Based Rendering and its current technical difficulties which are calculation of the accurate depth map and holes filling.In order to avoiding these technical difficulties,the theory and experimental research on this technology are carried out from two angles of sparse feature points and ray space in this paper.The specific research work and achievements are as follows:(1)A new method for generating virtual viewpoint based on sparse feature points is proposed.The method combines feature points,triangulation and image morphing to realize the generation of virtual viewpoint.In feature point extraction,the SIFT feature points algorithm is used;the optical flow method is introduced in order to improve the image quality of virtual viewpoint.After the corresponding relationship between the images is set up by matching the two part of the feature points,the Delaunay triangulation network is established by using the feature points as the vertices of the triangle.Finally,according to the image morphing theory,the virtual viewpoint of the reference image is generated by D3D.Experimental results show that the proposed method can achieve a better visual image quality.(2)A new method for generating virtual viewpoint based on ray space is proposed.The process of implementation of the method:Firstly,the multi-view images are converted to the data representation in the ray space.Then in order to form a dense ray space,the nearest neighbor method is used to interpolate in sparse ray space.Lastly,the dense ray space data is converted to the image space which is composed of the virtual view images and the real images.When large association between the front and rear frames in the collection of multi-view images,experimental results show that the proposed method can achieve a better visual image quality.