Research On Face Recognition Algorithm With Compressed Sensing

In recent years,face recognition technology develops faster and faster.It has been widely used in various fields of our daily life.Identification method based on face information has gotten more and more attention.Comparing with other biometrics technology,the face recognition has lots of advantages such as low cost,no invasion and difficult to be lost.So it has broad application prospects in security protection and financial payment fields.In the late years,many excellent face recognition algorithms have been proposed.But many problems exist in the practical application such as uneven illumination,the influence of occlusion,lack of samples and the changeful posture.So face recognition is still the research hot spot of many scholars.As the ability to getting data is enhancing in the multimedia electronics,the amount of data which must be dealt with in medical imaging and other aspects of the digital signal processing systems is increasing surprisingly.The traditional Nyquist theory has been hard to meet the requirements gradually.Recently,compressive sensing has become a new research hotspot in the field of digital signal processing.It breaks the limitation of the Nyquist sampling frequency and greatly saves storage space for sparse or compressible signal.The Compressed sensing theory were used by many scholars in face recognition and achieved some results.This paper finishes some improvement research on the basis of sparse representation and has achieved some good results.The main results and innovation points of this paper are as follows:(1)In order to solve the problem of the complicated background leading to mistakenly identified in collecting the non-restriction of face images,this paper adopts the Adaboost algorithm to complete the detection of Haar features and uses the feature points by AAM model to verify the accuracy.It reduces the face detection error rate.At the same time,the Adaboost algorithm is applied to detection of eyes to make the normalized processing of face effectively.It is proved that face detection is the important premise of face recognition algorithm in real application.Especially for patially occluded images,the face region should be detected to be normalized first.In this paper,the occluded face recognition algorithm based on SRC and overlap blocks is proposed.It gives full consideration to the whole face both the relevance of information and reduces the influence of the mask to the recognition rate effectively.Meanwhile,the robustness is enhanced.(2)In view of the samples' high dimension problems,a face recognition algorithm based on weighted wavelet transform and compressive sensing is proposed.Considering the influence of wavelet transform on each component,the low-frequency components,the horizontal and vertical components are weighted fused.It can reduce the loss of the identification information and the dimension of face images.Then,PCA(principal component analysis)is employed to extract facial feature matrices.Each class of the test samples is reconstructed by the sparse coefficients and the dictionary matrix.The test samples are classified in the smallest residual error by comparing with the reconstructed samples.Experimental results on the ORL and FERET face databases verify that the algorithm proposed in this paper has a better recognition ability.The recognition accuracy and recognition speed of the new algorithm are better than traditional methods.(3)According to the gender characteristics of face in each local area not consistent,this paper considers the algorithm based on local and the overall integration of gender identification.The face is divided into six sample subset;the subset of each sample is done sex identification respectively.Furthermore,on the basis of identification of a single sample is set by voting local samples of face recognition.Considering the complexity of calculation,sparse representation algorithm is introduced for redundancy characteristic in this paper.The sparse representation algorithm is used to reduce the common characteristics of redundancy and improve the computational efficiency.The human face of complementary information between different local is sufficiently used.integration.At last,the algorithm achieves good recognition effect on the CAS-PEAL database.