Research Of Gender Recognition Based On Face

Face is one of the most important features of human body, which can reflect a lot of important information, such as gender, age, race, identity properties. Gender, as a kind of important information, gains a widespread attention and becomes a promising means to make use of. Due to the virtue of its intuitive, Gender recognition based on face has already been the research hotspot in the field of biometrics.This thesis works on the research of face gender recognition thoroughly and systematically, and specific works as follows:(1)During the process of face detection, the number of Harr- like features is very large, so the cost of calculation is too much. Aiming at this problem, here puts forward Harr-like characteristic value calculation method based on the integral figure, and combined with Adaboost algorithm. After the procedure of training, the cascade classifier is generated, which has the ability to detect single face and multi-face in images. The cost of calculation is cut greatly, the speed of training Adaboost classifier and face detection turns out faster.(2)The dimension of face feature is too high to compute, and there exits image noise. aiming at these problems, Here puts forward a gender recognition method of combing the texture feature extraction method based on MB-LBP with Adaboost algorithm. By extracting the 2 * 2 block size LBP texture feature, the influence of most noises is eliminated. After resampling the texture characteristics of high-dimension, high-dimensional turns to low-dimensional. Based on the former procedure, here comes the training and generating process of the Adaboost classifier. Experiments show that the classifier is better than single pixel LBP features based on SVM algorithm.(3)As to single gender classifier, the function of classification is not very ideal. Aiming at this problem, here puts forward a method of cross combining the feature extraction algorithm, which including PCA, LBP, with the gender classification algorithm including Adaboost, SVM, SRC. By different combination of two type algorithms and experiments, several good forms of combination are picked out, then mix them together. Through the experiment, seven good classifiers are selected, then mix these classifiers together. Experiments prove that after fusion of gender classifiers, the classification works better than single gender classifier.