Research On Compressed Sensing And The Application For Biometric Technology

With the rapid progress of science and technology, biometric technology steps intothe peak of development. More and more biometric identification techniques areapplied in all the fields of economic and society, and provide technical support for thegeneration of security certified products. Because of this application, biometrictechnology is no longer the abstract of scientific research, but gradually becomes thekey power of development in the variety of fields and industry. The deep researchmakes biometric identification technique constantly serve our society and contributeto public. Meanwhile, if it were mastered that from image acquisition and recognitionto all kinds of research methods for identification algorithms, the key technique ofbiometric identification would be understood. It has also built a solid foundation forour related domestic industry and provision of international competitiveness.This report aim at the most representative Face Recognition and Finger VeinAuthentication in the biometric identification techniques. Combine with the very newtheory of compressed sensing, an image denoising algorithm on the venous based onthe sparse decomposition, which has a very good denoising effect for the finger veinimage. Because of there is a deficiency on the respect of independence and identitydistribution based on the two-dimensional linear discriminated analysis method forface feature extraction. We have studied the application of compressed sensingclassifier in the face recognition, and improved the face image recognition rate withthe two-dimensional linear discriminated analysis method. This report could be divided into five parts as followed:First of all, we introduced the literature research and its significance, meanwhile weelaborated the development status of biometric feature identification etc.Secondly, we thoroughly introduced three components of compressed sensing theory,which are sparse decomposition, design of observation matrix and signalreconstruction. We demonstrated how to reconstruct signal with matching pursuit,orthogonal matching pursuit and iterative hard threshold algorithm. Furthermore, weexamined an improved algorithm because of the deficiency of orthogonal matchingpursuit algorithm.Thirdly, we proposed a method of building atom library for the characteristics offinger vein image, which realized the finger vein image denoising based on sparsedecomposition. Comparing with the traditional denoising effect, sparsedecomposition denoising method is better.Fourthly, we have analyzed the most representative face recognition method in thebiometric feature identification, and introduced feature extraction method based ontensor which succeeded from traditional methods, and studied the two-dimensionallinear discriminated method of reorganization of the image sampling against thedeficiency usage of two-dimensional linear discriminated method in independentdistribution aspect.Finally, we studied the design method of compressed sensing classifier and faceimage identification effect. With the comparison experiment with traditional neighbor classification, we verified the superiority of compressed sensing classifier. On thisbasis, we did the face image identification experiment with two-dimensional lineardiscriminated feature extraction method based on the reorganization of the imagesampling, and verified the algorithm effect.This report and experiment are based on the project in the Laboratory Group–JilinScience and Technology Development Fund and National Natural ScienceFoundation. And part of research in this report has already published in domesticjournals, such as finger vein image denoising based on the sparse decomposition.