Adaptive IBR Technology Oriented To The Multi-view Stereoscopic Display

Image-based rending (IBR) is capable of generating multiple views of secenes realisticly and rapidly. So they are suitable for the image source acquirement technique of the multi-view stereoscopic disply. However, the over reliance on image sampling makes it a contradiction between the data quantity and rendering quality which limits the application of IBR.Centered on this problem, this paper derives and establishes the minimal expectation error criterion of the reconstruction of signal samples and presents an adaptive sampling and reconstruction method for multi-dimensional signals under this criterion. Baed on the analysis of the original signals, this method analytically determines the sampling positions which contribute the most to the reconstruction quality, thus improving the reconstruction quality and reducing the sample size. Because of the needlessness of extra scene geometry and the ability of adjusting the sampling rays individually, this method possesses the features of non-iterative sampling, low geometry knowledge requirement and individually adjusting the sampling rays which similar methods do not possess.When applying to the 1D (Audio), 2D (Image) and 4D (IBR) signals, an improved adaptive sampling method which combines brim and edge points is presented based on the basic method. It improves the reconstruction quality by reducing the large renconstruction error produced by the flat-roofed waves and"Sheet Iron Distortion". Specially, 4D Sobel kernel is derived to extend the edge detection to 4D spaces to facilitate this method to be used in IBR applications.The light field is used as the IBR realization in this paper. According to nonuniform data produced by adaptive sampling, Delaunay triangulation based nonuniform data interpolation is proposed and applied in light field reconstruction to effectively reduce the computation of the high-dimensional interpolation. Furthermore, the system structure is improved by buffering the trianglulation data in database to increase the efficiency of the system in case of wandering around the scene. Also a data transmission/storage structure appended by position labeling matrix is presented to implement the adaptive IBR system in practical conditions.Aiming at the requirements of multi-view stereoscopic display, this paper analyzes the effect of display parameters on IBR sampling and rendering parameters, and achieves adaptive IBR sampling, reconstruction and multi-view stereoscopic displaying. Indicated by the experiments, this method obtains 3.7~7.8db higher rendering quality than the traditional method with the same sample size, and acquires over 30db rendering quality with less than 15% of the original samples.From the perspective of application scope, this method is applicable to the light field as well as other kinds of IBR data and audio, video signals. From the perspective of the stage and the realization, this method is aiming at the data source and based on the presented adptive sampling method. It gives up the paradigm in this field which focuses on the post sampling stage and uses the existing video compress method. It provides a novel way for the IBR research.