[Paper] Compressed Sensing of Ray Space for Free Viewpoint Image (FVI) Generation

Free Viewpoint Image (FVI) is one of the most popular image format in the next generation of multimedia, and ray space is an effective and efficient method to generate FVI. However, in the traditional method of ray space construction, all the images have to be captured in advance so that the burden of data is quite heavy. In this paper, we propose to adopt compressed sensing to sparsely sense and reconstruct a ray space. Thus, it is not necessary to capture all the images but only fewer measurements are collected, and ray space can be reconstructed by employing optimization tools with sparsity promotion. Different from previous applications of compressed sensing in image acquisition, such as computational photography which is focusing on integral image, our work is based on Epipolar Plane Image (EPI). In our simulation, the ray space can be reconstructed successfully, and simulation results also illustrate the reconstruction performances from different numbers of measurement and different desired sparsities of EPI. Furthermore, since EPI presents unique structures, another dictionary which can represent this structure is also developed to take place of common orthonormal basis in compressed sensing procedure. Experimental results show that sparser representation of EPI can be achieved and better reconstruction can be obtained by using newly developed dictionary. Finally, the subjective testing results are also presented and FVI can be obtained from the reconstructed ray space.