Research Of Face Recognition Algorithm Based On Local Texture Features And DBN

Face recognition has always been a hot topic of identification.Feature extraction and classification recognition algorithms are two major problems in face recognition,but they are easily affected by factors such as expression,posture,accessories,hair,and lighting in non-specific situations.The paper focuses on the study of local texture features and deep belief network models of human faces.Main works are as follows:1.A face recognition algorithm model based on TPLBP feature and deep belief network was proposed.First,the model extracts the TPLBP features of the face;Then uses the deep belief network to learn and reduce the dimensions;Finally,the Softmax classifier is used to recognize and classify faces.The model uses TPLBP feature to overcome the problem of LBP features being susceptible to noise interference.The parameters of TPLBP and the number of nodes in the hidden layer of DBN were determined through analysis and experiments,and finally a good recognition effect was obtained.2.A weighted local ternary pattern(WLTP)algorithm is proposed,and a new face recognition algorithm is presented in combination with the depth belief network.This algorithm obtains the WLTP feature of the face and uses PCA to pre-dimensional reduction,and then learns and recognizes it by the deep belief network and Softmax.WLTP algorithm assigns different weights to the contribution of recognition in different regions of the face,which not only exerts the robustness of LTP algorithm to lighting,but also gives reasonable weight to different regions,and improves the recognition efficiency.3.Based on CMUPIE face database and FERET face database,two kinds of face recognition algorithms have been trained and tested,the influence of factors such as parameters,characteristics,data set size,segmentation,and number of network nodes on recognition rate and speed was discussed,and experimental data were given.