Parallel view synthesis programming for free viewpoint television

View synthesis is one of the important techniques for free-viewpoint 3D image services. Unfortunately, computational complexity of depth image-based view synthesis is high, since it includes numerous matrix calculations and complex filters. In this paper, we implement the depth image-based rendering algorithm on a graphics processing unit (GPU) using the compute unified device architecture (CUDA). We perform memory uploads to the global memory of the GPU, and compute matrix calculations of all pixels in parallel. We also simplify the filters of rendering to reduce complexity and analyze performance according to block and image sizes. Experimental results show that our implementation is faster than conventional methods while preserving visual quality.