Research And Implementation Of Iris Recognition Algorithm Based On Image Region Segmentation

With the need of military reform and social development,because of the characteristics of non-invasion,stability as well as uniqueness,iris recognition technology was widely used in the modern military,national security and civil fields.The whole iris recognition system was based on four parts:iris image sampling,image preprocessing,feature extraction,image matching and recognition.Iris localization and texture feature extraction were the key technologies of iris recognition system.This thesis was mainly as below.(1)Iris image preprocessing.First,the coarse location of iris inner boundary was realized with the method combining morphology with connected component labeling.And then fine location of iris inner boundary was achieved by the combination of edge extraction and small-range searching method.Outer boundaries of the iris were located with edge extraction,denoising method,and the small range search method.Then,the positioned iris image was first denoised,and then converted to a rectangle image in size of 48×256 with polar coordinate conversation.Finally,by image enhancement processing using histogram equalization,the iris normalization image was gotten.The 756 images in CASIA version 1.0 iris database were preprocessed.Only 10 images were located failure,and accurate localization rate was reached 98.7%,and the average location time was 1.9365 seconds.(2)Feature extraction and pattern matching of Iris image.The thesis adopted algorithm of two-dimension optical Haar wavelet transformation based on image region segmentation to extract feature.First,the whole iris normalization image was decomposed with 3 layers of decomposition with Haar wavelet,and the third layers of diagonal detail was selected.Then,the iris normalization image was divided into 8 blocks,and each blocks was decomposed with 3 layers of decomposition,and horizontal and vertical details were alternately selected.Finally,the 9 details were coded to build an iris feature template,which was in size of 384 bits and was classified and recognized with Hamming distance.According to experiment,the correct recognition rate in this algorithm was 96.3%.(3)The algorithm of two-dimension optical Haar wavelet transformation based on image region segmentation was improved.First,the image was divided into left and right parts.Because of the upper and lower eyelids which had been eliminated brought about the correlation of each iris normalization image was larger,the upper eyelid was removed effectively,through the image regions divided finely.Finally,the 9 details were coded to build an iris feature template,which was in size of 240 bits.According to experiment,the correct recognition rate in this algorithm was 97.38%.In addition,the thesis also optimized the improved algorithm,and the regions were divided more finely and the upper and lower eyelids were removed more effectively.According to the experiments,the correct recognition rate in this algorithm was 97.68%and the average time was 0.049656 seconds.