LUT filters for quantized processing of signals

We introduce a method to perform filtering on approximations (quantized versions) of the input signal, which lends itself to a practical implementation solely based on look-up tables (LUTs). The LUT filter approximates the performance of some traditional nonlinear filters at a fraction of the cost. The filter is divided into an approximation stage that is constant for all filters and a filtering stage, which is one LUT that can change in order to implement different filters. We introduce an overlapped hierarchical vector quantization (OHVQ) scheme that is used as the approximation stage. The output is produced by mapping the OHVQ codes to filtered data. Hence, all processing is done via LUTs, even though the filter size needs to be small because of typical OHVQ contraints. Switching among filters demands changing pointers to only one small LUT. Preliminary analysis and image processing examples are shown, demonstrating the efficacy of the proposed method.