Reduced dimension image compression and its applications

A key to successful image compression is a combination of (1) energy compaction by appropriate transform algorithms to exploit data redundancy, (2) judicious quantization of compressed transform coefficients, and (3) efficient entropy encoding that takes advantage of different symbol rates. The reduced dimension image compression (ReDIC) algorithm consists of (1) a library of transform algorithms to exploit data redundancy in local subimage, (2) subspace filtering to maximally decorrelate transform coefficients over multiple subimages and to achieve additional dimension reduction to overcome the curse of dimensionality, (3) vector quantization (VQ) of compressed transform coefficients, and (4) entropy encoding of VQ indexes with Huffman or arithmetic coders. We describe the ReDIC algorithm and discuss its application to image coding of still pictures and sonar spectrograms for low-latency, low-cost wireless or satellite transmission to shore processing stations. We quantify the impact of image compression on minimum detectable lines with deflection or detection index. We also compare the original and reconstructed spectrograms derived from real data at a 52:1 compression ratio. We also discuss how image coding can be applied to pattern recognition.