Identity Recognition Based On Deep Face Information

Face recognition technology has made great progress after years of development,especially under controlled conditions,the performance of the algorithm is more significant.However,most of the recognition algorithms only stay in the surface information of the face image,and can not extract the deep information hidden by the face image itself.Therefore,when the environmental factors are uncontrollable,the performance of the face recognition algorithm is often seriously threatened.Common influencing factors include illumination,attitude,occlusion,etc.These factors often restrict the application of face recognition algorithms.Therefore,studying how to extract face image depth information not only has academic value,but also has certain application value.An improved local ternary pattern(LTP)feature descriptor is proposed to improve the extraction efficiency of facial features.The improved adaptive threshold algorithm is used to improve the performance of the operator's description of local features,and in the histogram representation process of the feature vector,the feature representation ability of the face image is improved according to the local weight of the image.Through the comparison experiments on Yale-B,AR,CMU-PIE face database,the effectiveness of the proposed operator is proved,but it also exposes the shortcomings of recognition accuracy in complex uncontrolled environment.Based on the proposed local feature descriptors,a depth-feature extraction method based on multi-scale central symmetric adaptive local ternary mode and convolutional neural network(CNN)is proposed.The method makes full use of the extracted robust local texture characteristics to illumination and occlusion.It compensates for the insufficiency of the CNN to completely extract more discriminative facial features when directly using the original image.At the same time,the method contains high-level abstract features in different directions,different scales and depths,and the face information is described in more detail.By comparing experiments in the LFW face database,the performance of the algorithm is improved to some extent.Finally,this paper uses the above-mentioned deep face information algorithm to construct a face recognition system that can complete the face recognition process and obtain face identity information.