Case study of binary and ternary synthetic discriminant function filters with similar in-class and out-of-class images
A simulation study is reported of the performance of two designs of ternary synthetic discriminant function (SDF) filters. The filters are compared to binary SDFs in terms of the probability of correct identification P(ID) between in-class and out-of-class image sets. Compared to binary synthetic discriminant function (BSDF) filters, both ternary SDF filters designs offer improved performance in P(ID) and thus allow extension of the maximum filter direction range when evaluated in the presence of white image noise. Larger distortion range filters translate to fewer correlations required to perform image identification, and thus higher system speed. The TSDF(SNR) filters had the highest of system performance, while TSDF(d) filters were able to improve discrimination to a degree, but their optimal designs had essentially the same SNR performance as the BSDF filters.