A New Tabular Algorithm for Rapid Vector Quantization Encoding and Nearest Neighbor Classification

We offer an approach to a multiple-order-of-magnitude processing speed/efficiency improvement in vector quantization (VQ) data/image compression, automatic target recognition/classification, and time signal/signature classification. We base our approach on an algorithmic breakthrough, the TABULAR NEAREST-NEIGHBOR ENCODER (TNE), which we discovered and partially developed last year. This algorithm reduces the computational complexity associated with such problems from 1 to over 4 orders-of magnitude compared to other techniques. A significant property is the relative insensitivity of the algorithm's complexity to codebook and/or vector partition size. This property renders feasible the use of very large image quantization or feature vector codebooks in real-time with relatively modest processor hardware. Larger codebooks will render practicable larger vector partitionings and thence greater data compression ratios than were previously associated with VQ methods. The TNE architecture is well suited for parallel processing and for independent sub-space (i.e., correlated partial image block) classifications. This latter property will greatly facilitate model-based classification of partially obscured target imagery and partially masked time signatures. This effort will significantly advance the state-of-the-art in real-time image/data compression; in automatic recognition/classification of exposed and/or partially obscured targets; and in related commercial applications (multimedia, video compression, speech recognition/compression/synthesis, etc.)