Evolutionary reconfigurable architecture for robust face recognition

This paper proposes a novel reconfigurable architecture with capability of evolution/adaptation, called ERM (evolutionary reconfigurable machine), and it is implemented partially on FPGA chip. Evolutionary module which has been implemented by parallel genetic algorithm evolves filter blocks, and feature space to achieve an optimal face recognition configuration of the ERM. Since a priori information of noise and system working environment are not available, heuristic intuitive decisions or time-consuming recursive calculations are usually required. The ERM can explore optimal configuration of filter combination, associated parameters, and structure of feature space adoptively to unknown illumination and noisy environments. Some of the commonly used filters such as median filter, histogram equalization filter, homomorphic filter and illumination compensation filter are designed and verified by implementing on FPGA hardware. Parallel genetic algorithm evolves the connection and parameters of image enhancement filters as well as feature space of Gabor representation. The ERM has been tested to the face recognition in varying environments of illumination and noise patterns. The varying illumination environments include light direction, contrast, brightness, and spectral composition. The proposed architecture for face recognition adapts itself to varying illumination and noisy environments using the evolutionary computing method. The experiment performed using the ERM shows very encouraging result, especially for changing illumination and noisy environments.