Real-Time Super-Resolved Free-Viewpoint Image Synthesis from an Array of 25 Cameras
Details
This paper presents a system that performs real-time super-resolved free-viewpoint image synthesis, in which, from a given set of low-resolution multi-view images captured by a camera array, a free-viewpoint image is synthesized with a higher resolution in real-time. The process of the synthesis consists of two steps: depth estimation with a modified semi-global stereo matching, and reconstruction-based super-resolution. We implemented this process in graphics processor unit (GPU) for speeding up. Our system is capable of synthesizing a high-resolution (320×240 pixels) free-viewpoint image from 9-view low-resolution (160×120 pixels) around 40 ms.
