Nearly Lossless Compression of Arbitrary Images

摘要

Multiple videos such as those collected using monitoring and surveillance applications generate large amounts of data. Significant reduction in data-rates are achievable if frames in these video sequences are jointly compressed. This paper describes a compression algorithm that can jointly, pair-wise compress two (or more) video frames by creating one video frame and a difference frame. The difference frame is computed by taking the difference between DCT coefficient matrix of the two frames. No similarity between frames or moving objects is assumed. Instead, we rely upon the inter-frame and intra-frame correlation characteristics at the macro-block level. The technique is applied for frames exhibiting inter-frame correlation coefficient as low as 0.1. A compression-ratio of close to 30% is obtained.