Comparative Study Of Compressive Sensing-based Approaches For Face Recognition

Compressed sensing(CS)theory has been an active research topic in signal processing field in recent years,and has penetrated into mathematics,telecommunications and many engineering fields.As for face recognition,it has been widely applied in security monitoring,identity authentication,entry and exit administration,crime picture retrieval and so on because of its conciseness,noninvasive and low cost.The sparse representation classification(SRC)is a more efficient face recognition approaches with face images have occlusion and corruption compared with traditional algorithms.However,SRC algorithm is not merely sensitive to alignment,but also time-consuming.This thesis focus on fundamental notation based compressed sensing,how to improve the robustness of existing algorithms and reduce time consumption.The main research works are summarized as below:1.Some face recognition algorithms sparse representation-based were summarized,such as SRC and collaborative representation classification.An improved classifier which has much better efficiency was proposed.2.In order to address the challenges that face images in high dimension waste too much calculation resources,a projection optimization method(POM)which based on the theory of CS was proposed,then this method was used as feature extraction method for dimension reduction.A comparative study was present between our algorithm and some traditional feature extraction method.And we prove the feasibility by experiment of face recognition.3.In order to address the challenges that SRC algorithm didn't concern about the local information of training face samples,an algorithm based block dictionary learning was proposed.SRC algorithm uses all training face samples to structure a unified dictionary and looks for the sparest representation of a test examples in a dictionary composed of all classes.In order to deal with this situation,we propose an algorithm to attain optimal block dictionary by K-SVD with block faces images as initial dictionary,for each person an independent dictionary was trained for classify.The experimental results exhibited that the proposed algorithm can achieve much more satisfactory robustness compared to other algorithms.