Face Recognition Based On Feature Learning In Shearlet Domain

As the face image has the advantages of fast,intuitive,non-contact,accurate and reliable;since the human activities,the face image is regarded as a way to identify the identity.Face recognition technology has been applied in the field of national defense security,criminal detection,business security and other fields,is an important research field.Wavelet transform has good time-frequency analysis capability and has been widely used in face recognition.Shearlet transform is a new multi-scale geometric analysis tool in recent years.With the traditional multi-resolution and time-frequency local characteristics of wavelet transform,it has overcome the shortcomings of traditional wavelet transform.The sparse representation of the image has a higher computational efficiency.Depth learning by learning a non-linear network structure,to achieve the input data abstract expression,from the construction of network structure for face recognition.Based on the feature acquisition of face images and how to classify the face images,this paper studies two kinds of face recognition algorithms to improve the recognition results.The first is based on shearlet transform is proposed based on Nonsubsampled Shearlet Transform(NSST)on the face image characteristics of sparse representation and Deep Belief Network(DBN)algorithm for face recognition.Innovation points:(1)shearlet due to the sparse representation feature in shearlet domain is proposed to extract features,can effectively capture the face image edge information and texture information of face images get rich;(2)taking into account the effect of feature extraction on the recognition results,propose a new block weighted by LBP algorithm.Extraction block weighted LBP feature in shearlet domain,in order to obtain better feature representation of face image;(3)based on automatic feature learning,feature learning and classification and selection of deep belief networks,in order to weaken the characteristics of impact on the recognition results,and constantly improve the recognition results.The experimental results show that the face recognition algorithm based on non sampling shear wave and depth belief network has good effectiveness and robustness.The second work is to propose a face recognition algorithm based on deep-level feature and sparse representation.Innovation points:(1)for face recognition was improved,put forward in deep-level feature to compensate for the lack of feature extraction of original features can only be fixed in the correspondingmode of optimization;(2)to feature sparse structure,increase the diversity characteristics;(3)the deep-level feature and sparse representation.With proposed face recognition algorithm based on the characteristics of deep learning and sparse representation,and the performance of face recognition algorithms are analyzed.Compared with the current face recognition algorithm,the applicability and efficiency of the face recognition algorithm based on representation feature and sparse representation are explained.