Construction of Face-Sample Set by Applying Genetic Algorithms

Chen Jie 1 Gao Wen 1,2 Vilab, Computer College, Harbin Institute of Technology, Harbin, China, 150001 [email protected] 2 Institute of Computing Technology, CAS, Beijing, China, 100080 [email protected] Abstract: This paper presents a novel construction of face-sample set method by applying genetic algorithms (GA). The GA can yield new face samples by crossover and mutation. These new samples will have some new prosperity: simulating the variations of face in daily life and the variations of the image of light and quality. First we align the face samples with one another to reducing the amount of variation between images of faces and improve fitness of solutions. And then these face samples are divided into three sub-sets: train set, validation set and test set. The train set is then used to train a Sparse Network of Winnow (SNoW). And the train set is also used as the initial population of GA. After every 20-generation, we will use the initial generation and solutions with high weight, evaluated by SNoW, to train the network SNoW again. And each newly-trained SNoW is tested by validation set. Finally, these trained classifiers are checked by the test set to check its generalization performance. The experiment results prove the new train set can improve the performance of SNoW obviously than only the original train set.