Research On Face Recognition Based On Sparse Representation

With the development of society, it has became a popular way in securityapplications to identify personal identity quickly based on biological feature, and ithas gradually became a hot research topic at home and abroad.Because of theinternal property of biological characteristics it has heavily otherness and stability indifferent people.Face recognition is an important technology in biometriccharacteristics recognitions.It is more directly,friendly and conveniently to otherbiometric features recognitions, and it is easy for the user to accept. At the same time,pattern recognition and image processing etc have been improved by the research offace recognition. In brief, there is a wonderful application prospect for facerecognition.Face recognition based on sparse representation is a kind of new way whichusing compressive sensing to face recognition. In my research,face recognition basedon sparse representation will be researched in-deepth and distributed compressdsensing will be used to face recognition.On this basis,the natural-light problem anddimension reduction problem will also be researched. The main research results areas follows:(1) A way of face recognition based on Gabor filter and joint sparse model is putforward. In this new algorithm, first, the original face image must be dealed withGabor filter then the Gabor characteristics can be required. Second, extract the publicand private characteristics of training images of each class by the way of joint sparsitymodel. At last, the extracted characteristics are used to reconstruct the test sample andthen determine the class according to the reconstruction error. Experimental resultsshow that this new algorithm is robust to light and the storage space is saved.(2) A way of combining the joint sparsity model and sparse representationclassification is put forward.Because of joint sparsity model is not only consideringthe signal correlation within the class but also the correlation between classes,and thestorage space is also saved. At the same time, sparse representation classification hasan accurately classification function. Consequently, in this new algorithm, the publicparts and pravite parts of each class are abstracted to structure dictionary and then thesparse representation classification is used to determine the class. Compared to thesparse representation classification, the itom number of dictionary is decreased. (3) A way of face recognition based on joint sparsity projectionis is put forward.The cost of count can be reduced effectively by dimension reduction, and therecognition effect can be influenced by dimension reduction matrix. So in thisresearch, on the basis of joint sparsity model, the public and private parts of theimages are used to reconstruct the orginal images, then according to solving theoptimization problem to deduce the dimension reduction matrix. Thus it ensure theimage in the low-dimensional space still retaining the main features in thehigh-dimensional space.On this basis,using the dimension reduction matrix to reducethe dimension of image then determine the class.The threes ways in this research are respectively tested in the database of Yaleand CMU AMP.we also analysis the results. All of these results show a betterrecogination rate, and the feasibility and effectiveness are also proved.