Reduced dimension image compression for remotely distributed underwater signal processing

A key to successful image compression is a combination of (1) energy compaction by appropriate transform algorithms to exploit data redundancy and (2) efficient entropy encoding of transform coefficients to take advantage of different symbol rates. Further, it is important that one selects the transform algorithm that most efficiently addresses the system requirement. 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, (3) vector quantization (VQ) of compressed transform coefficients, and (4) entropy encoding of VQ indexes with Huffman or arithmetic coders. The intermediate step of subspace filtering has an added benefit of mitigating the effects of the so-called curse of dimensionality in VQ. The authors describe the ReDIC algorithm and discuss its application to image coding of sonar spectrograms for low-latency, low-cost wireless or satellite transmission to shore processing stations. They quantify the impact of image compression on minimum detectable lines with deflection or detection index. They also compare the original and reconstructed spectrograms derived from real data at 52:1 compression ratio