Probabilistic neural networks structure optimization through genetic algorithms for recognizing faces under illumination variations

Recognizing a face is very challenging in many applications, however, the process of converge to a conclusion of a known-face based from a 2D incoming face images is very difficult. It is well known that the performance of automatic face recognition system decreases significantly when large illumination variations are present in input space. In this paper, we implemented the illumination compensation praprocessing system in conjuction with the optimized-Probabilistic Neural Networks as a classifier. PNN has shown marvelous higher recognition capability with high speed of convergence, compare with that of low speed convergence of Back-Propagation neural system. Optimization of PNN is accomplished by determining the best smoothing parameter using Genetic Algorithm and the most importance neurons component through orthogonal algorithm. Experimentson on face recogniti are conducted using face images under various illumination conditions. Results show that the optimized-PNN using illumination compensation processes could achieve high recognition rate with low computational cost.