Face Recognition Based On Sparse Coding

Because of its friendliness,non-invasiveness and easy access,face recognition has more extensive applications than other existing biometric technologies,such as information security,human-computer interaction and entertainment.However,it is difficult and important points in face recognition research to extract face features with good distinguish ability and good robustness and construct efficient and reliable classifier to improve the correct rate of face recognition:.Face detection,feature extraction and face recognition are the three components of the face recognition system.Their quality directly determines the performance of the algorithm.The status of feature extraction is particularly prominent.The paper takes face recognition with single-sample as the background,and related research results based on feature extraction method,dictionary learning method and collaborative representation classification algorithm.Based on sparse coding method,relevant research is done.The main contents of the work are as follow:.(1)When the changes of external illumination are Gaussian or Laplacian,the Collaborative Representation Based Classification(CRC)can better overcome the effects of illumination.However,the actual illumination changes do not meeting this premise.Therefore,in order to overcome the influence of external illumination changes,a face recognition algorithm based on orthogonal Log-Gabor filter and collaborative representation is proposed.The orthogonal Log-Gabor filter is an improvement of the Log-Gabor filter.The improved filter not only retains the good biological characteristics of the Log-Gabor feature,but also alleviates the influence of the high dimension of the Log-Gabor feature.At the same time,the algorithm is also a combination of Log-Gabor and CRC.Therefore,it not only retains the advantages of the CRC algorithm,but also mitigates the effect of illumination changes on the CRC.Experiments show that the algorithm can deal with illumination problems better under the condition of face recognition with single-sample.(2)Face recognition with single sample is the most common in practical application scenarios.To overcome the small sample problem,an embedded relaxation collaborative representation face recognition algorithm based on multi-scale differential excitation is proposed.The multi-scale differential excitation model is an improvement of the differential excitation model,which aims to obtain discriminative features with more discriminative power.In order to obtain a more distinguishing coding coefficient,the algorithm combines the advantages of the relaxed collaborative representation model,and proposes an embedded relaxed collaborative representation model to make the linear representation of the test sample in the dictionary more accurate.Since there is only one training sample,in order to obtain more recognition features,the algorithm introduces a block-overcoming mechanism to increase the number of sample.By running the proposed algorithm on each sub-block of the test sample,the classification is achieved based on the classification criteria of minimum error.The experimental results show that in the context of face recognition with single sample,the algorithm not only has a higher recognition rate for the face of illumination,expression,occlusion and attitude change,but also the running time of the algorithm is smaller.