Quantitative comparison of lossless video compression for multi-camera stereo and view interpolation applications

Computational video multi-camera systems allow novel applications such as stereo-vision and view interpolation. The computational- as well as communication and storage requirements for real-time multi-camera video are huge. High quality stereo- and view interpolation applications require the accurate combination of detailed image features in two or more cameras. The use of lossy video compression algorithms often lowers the accuracy of small details and textures that are probably not noticeable by a human viewer, but that are crucial in disparity calculations, matching, video stitching and 3D model synthesis. This paper makes a quantitative comparison of two lossless video compression methods. The intention is to use them for efficient implementation in System-on-Chip (SoC) architectures in computational camera systems. The methods compared are based on predictive-corrective compression and Huffman encoding as well as derived methods. For efficient hardware implementation alternative methods for the use of the Huffman coding are investigated. The comparison includes the use of Huffman encoding parameters from previous frames in the compression of current frames.