View synthesis for lightfield displays using region based non-linear image warping

In this paper, we present a view synthesis method for generating multiview image sequences for 3DTV systems using a sparse set of views obtained from cameras in a multiview linear camera configuration. First, the input images are analyzed to extract information about sparse disparity and a mesh is constructed using a set of sparse disparities on all virtual views. Then, for each virtual view, pixels are interpolated inside the mesh by formulating and solving special warping functions and by fitting a uniform bicubic surface to the original data points from the input images. The method is fully automatic and can generate visually pleasing virtual views. Furthermore, we do not need any post processing operations like occlusion handling, hole filling or inpainting because of the warp driven approach. The method also supports view extrapolation in a limited range and can be implemented in real time, which is extremely needed in the present scenario.