Multimedia Data Compression Based on Pattern Matching

Principal Investigators: Mikhail Atallah, Wojciech Szpankowski

Now days we must handle large amounts of multimedia data such as text, image, video, and audio. It should be clear that multimedia data compression is indispensable for various applications. In order to treat uniformly all the media, a new data compression scheme is required Since most media are represented in a digital form, it is natural to apply a technique that was proved successful for inherently digital media such as text, namely Lempel-Ziv type schemes. We shall propose such a scheme that works for all media by extending lossless Lempel-Ziv to a lossy scheme based on approximate pattern matching. Experimental results for Pattern Matching Image Compression (PMIC) showed great potential for such a scheme. The following features of this new homogeneous pattern matching compression scheme should appeal to various applications: (i) theory (there is an information-theoretical explanation why this scheme should work well, which suggests that it may yield more reliable performance than heuristic/ad hoc approaches); (ii) fast/efficient decompression (this scheme does not require powerful computing resources on the decompression end); (iii) good handling of high frequencies (for image compression the scheme does not blur the high frequency information such as edges, etc.); (iv) uniform approach to multimedia (as mentioned above, by re-defining distortion measure we can use this approach to put together a uniform approach to all media such as text, image, video, sound, etc.).