PSO Optimization of MLP Face Detection System with Min-Max Features

Face detection has wide applications in biometrics verification and interactive devices that employ human-computer interfaces. The face detection method presented in this book creates a robust yet compact classifier that can be rapidly retrained due to its structure. A new set of features called New Min-Max Analysis (NMMX) that derive horizontal and vertical projections from segmented grayscale image regions. and extracts vertical and horizontal projections from them. Then, a modification of a swarm-based optimization algorithm called Modified Particle Swarm Optimization 1 (MPSO1) was used to select the three most discriminative features from NMMX. To maximize the classification accuracy of the MPSO1-optimized MLP, another modified version of PSO called Modified Particle Swarm Optimization II (MPSO2) was used to optimize the structure of the MLP classifier. The final model achieved significant structural reduction compared to the benchmark system, while managing to significantly improve classification rates. The MLP was also tested on a set of real-life images that represent the face detection realistic scenario with detection rates of 88.17%.