Multi-objective adaptive composite filters for object recognition

摘要

An iterative algorithm to design optimal trade-off correlation filters for pattern recognition is presented. The algorithm is based on a heuristic optimization of several conflicting metrics simultaneously. By the use of the heuristic algorithm the impulse response of a conventional composite filter is iteratively synthesized until an optimal trade-off of the considered quality metrics is obtained. Computer simulation results obtained with the proposed filters are provided and analyzed in terms of recognition quality measures in cluttered, and geometrically distorted input test scenes.