Spatial Mask And Diffusion Filtering In Surveillance Video Compression

A surveillance-centric video compression algorithm is discussed that exploits a background model, motion estimation, truncated difference correction, and entropy encoding. The algorithm\u27s architecture allows tradeoffs between image quality and compression to target regions of salient activity. A set of window-based filters and heat diffusion PDEs is examined for impact on compression ratio and signal quality. Results show that filtering techniques are effective at reducing certain contributions to the data stream with minimal impact on image quality. Results from other compression codecs are included for comparison. The test set comprises a diverse range of surveillance scenes featuring vehicular and pedestrian traffic. © 2012 SPIE