Real-Time High-Fidelity Compression for Extremely High Frame Rate Video Cameras

Abstract

The single most debilitating bottleneck for sustained capturing of ultrahigh speed high-resolution video is the limited bandwidth of the on-camera mass storage device. We propose a novel downsampling-based real-time compression algorithms together with a suited camera architecture to overcome the bandwidth problem. The encoder generates and embeds into the compression code stream some useful side information to assist the decompression at the decoder end. The fidelity of recovered videos is made scalable to the decoder complexity, ranging from real-time decoding for instant playback at lesser quality to offline high-quality reconstruction by sophisticated sparsity and learning-based recovery algorithms, which are aided by the side information.