Research Of Iris Recognition Based On Sparse Representation And Collaborative Representation

Many of the acquired iris images suffer from the occlusion due to the specularreflections, eyelashes and eyelids, which affects the performance of the system.Recently, sparse representation-based iris recognition algorithm was introduced, and itpossesses strong robustness and effectiveness in overcoming the occlusion. Withregard to the above issues, the thesis conducts in-depth study on sparserepresentation-based iris recognition on the basis of related research achievementshome and abroad and the latest research progress.Firstly, the iris image is divided into blocks, and an improved block method isproposed which is along with the angular and radial direction. The local features areused to recognize separately based on sparse representation and the recognition resultsof different blocks are combined by Bayesian fusion to get the final recognition result.Simulation results show that the performance of the proposed algorithm has bettereffectiveness and robustness than the traditional iris recognition whether in theauthentication or recognition mode.Secondly, the sparse representation ignored an actual situation that most of thesparse coefficients have the ‘block-sparse’ structure. Coding the block structuresparsity can effectively reduce the degree of freedom in the sparse coefficients andimprove the reconstruction function for the sparse signal. Iris recognition algorithmbased on the block structure sparsity is proposed in this paper. One class of the irisimages is regarded as a block to use the correlation between different classes, whiletake advantage of the local features. Experimental result shows that the proposedmethod is superior to the sparse representation of iris recognition.At last, the recognition rate based on the sparse representation and block structuresparse representation will drop dramatically when the available training samples persubject are very limited, and the computational cost is high. To solve this problem, irisrecognition is operating collaborative representation on multi-scale patches andcombining the recognition outputs of all patches. Instead of recognition the entire iris image directly, the iris image is divided into several non-overlapping patches with thesame scale. Considering the fact that patches on different scales could havecomplementary information for classification, iris images are patched on multi-scale.The different multi-scale patches are recognized separately based collaborativerepresentation which reduces the computational complexity, while the ensemble ofmulti-scale outputs is achieved by Bayesian fusion. Experimental results on irisdatabases show that, although both training and testing image per subject might bevery limited, the proposed method outperforms the state-of-the-art recognitionapproaches on effectiveness and computational cost.