Abstract

We present a fast GPU (graphic processor unit) implementation of super-resolved free-viewpoint image synthesis using lower-resolution multiview images. To speed up the super-resolution reconstruction process, we propose an efficient and compact matrix representation that is suitable for parallelization using a GPU. We also modified the initial image and regularizer for the reconstruction process to reduce the iteration time until convergence. As a result, we achieved a speed up of about 70 times compared to the original CPU-implemented algorithm, requiring only 130 ms for synthesizing a 320×240 pixel free-viewpoint image from 160×120 pixel input images.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call