Optimal entropy encoders for mining multiply resolved data

A prototype client-server implementation of image analysis and compression is described which is based on the recently developed theory of Cohen, Dahmen, DeVore, and Daubechies for optimal entropy data encoders. The class of algorithms resulting from this theory was developed for the analysis and synthesis of data and yields optimal (in an information-theoretic sense), progressive, universal encoders for purposes of compression, storage, and transmission of data which can be developed into a multi-resolution framework. Such data include photographic and sensor images, digital terrain maps, and multidimensional scientific data generated by computational simulators. Two versions of the tree encoder have been implemented with a common client interface in order to demonstrate the applicability to diverse model problems. The first version processes the raw data in real time and is designed for relatively small data sets on single machines. The second version is designed as a network application to efficiently navigate archived data residing on remote databanks. Data structures which contain summary statistics, indexing and threading information generated by the analysis stage of the algorithm are utilized to navigate the data at adaptively-determined levels of resolution based upon the local depth of detail requested by the client application.

Optimal entropy encoders for mining multiply resolved data Conference Paper