Reduced cross-discrimination for discriminative filters

Image template detection is usually a very important halfway step for a computational vision algorithm. A recently proposed approach for it involves the use of discriminative filters. For a given template, they maximize the energy concentration in a single sample of its output. In this paper, we propose an extension for them that allows multiple template discrimination with reduced cross discrimination, using impulse restoration techniques. Simulation results suggest that the proposed method performs well, being an effective tool for decision problems.