Research On Face Detection And Gender Recognition Based On Convolution Neural Network

Convolutional neural network is a special kind of deep neural network model, which is a new type of artificial neural networks and artificial neural network technology combined with deep learning produced with local experience area, hierarchy, feature extraction and classification training process combined with global characteristics, in the field of image recognition has been widely applied. Particularity reflects convolution neural networks in two ways, on the one hand it’s a non-connection between neurons fully connected, on the other hand the same layer some right connections between neurons weight is shared, this non-full connectivity and weights shared network architecture reduces the complexity of the network model, reducing the number of weights, the network structure of the translation, rotation, tilt, scaling and other highly invariant.This article is designed to detect highly variable can be robust face detection algorithm based on convolution neural network structure of face images automatically from a group of people face and no face images in the training set, the automatic synthesis of specific issues simple feature extraction, without the need to make any assumptions or use of any manual feature extraction and face area. Convolution neural network structure and training methods are designed for rotation, occlusion complex background image with robust between detection rate and false alarm rate achieved good balance. Experimental results show that an effective face detection system before the image region classification does not require any expensive local pretreatment. Thesis algorithms focus on three challenging to achieve very high data rate of face detection, particularly low false alarm rate, without the need to use multiple networks to handle difficult situations.We also designed a gender-based human face recognition algorithm convolution neural network structure, direct input face image detection obtained, avoiding the complex image pre preconditioning. Feature extraction and pattern classification simultaneously, convolution layer and the lower layer composed of a plurality of sampling characteristic graph composition, and may be obtained through the learning and training. Through experiments on different face databases studied whether factors convolution neural network illumination, rotation, occlusion affect human face robust. The effect of the filter on network performance, by increasing and decreasing the number of filters to create three models applied to data sets experiments and found that the number of filters affect the classification ability of the network to select a suitable filter still need to keep the number of experiments to artificial selection.