VIEW SYNTHESIS OF UNCALIBRATED STEREO IMAGES USING FEATURE-BASED METHOD AND SINGULAR VALUE DECOMPOSITION

摘要

View synthesis has been an active area of research among the computer vision and computer graphics community due to its fast and less complicated rendering of novel views as compared to the conventional 3-D reconstruction-projection procedure. In this paper, we will present an algorithm for rendering a novel view of object from two uncalibrated stereo images using a sparse set of features and utilizing singular value decomposition (SVD) for correspondence matching. This algorithm has the advantage that no knowledge of the intrinsic and extrinsic camera parameters is required. The weak calibration which is represented by the epipolar geometry is sufficient for the generation of novel views. We will also present results of the algorithm on real and synthetic images.