Depth Estimation and View Synthesis for Immersive Media
Takanori Senoh, Nobuji Tetsutani, Hiroshi Yasuda · 2018
Light field images and multi-view images are a promising immersive media, which realizes naked-eye three-dimensional scenes or 6 Degree of Freedom (6DoF) Free Navigations. Since they require a huge number of pixels for a rich immersive media, an efficient data compression method is essential. Since the light field images are compatible to the multi-view images, we are investigating a decimated multi-view coding method using MPEG Depth Estimation Reference Software (DERS) and View Synthesis Reference Software (VSRS). In this paper, we analyze Homography projection in DERS and VSRS and propose their improvements. We also explain the depth estimation and view synthesis process in detail and report an experimental result with the improved software (pDERS8 and pVSRS4.3).