Efficient Algorithms For Iris Pattern Recognition

Biometric identification is a comprehensive technology to identify individuals according to their physiological as well as behavioral characteristics which ensure the high level of safety.There are different biometric-identification techniques that are being used in different human i-dentification systems such as iris recognition,face recognition,fingerprint recognition and voice recognition etc.Among all these real-world identification techniques,iris recognition is deemed most effective and reliable biometric identification technology because of its static and unique pattern features.Efficient iris recognition method refers to automatic identification and verifica-tion of a match between different human irises.Iris recognition system is generally composed of four main steps:iris segmentation,iris normalization,feature extraction and matching.In this dissertation,three efficient algorithms for iris identification are developed to im-prove the performance of existing iris recognition systems.The performance of the proposed algorithms is evaluated by taking into consideration the False Acceptance Rate(FAR),False Rejection Rate(FRR)and Equal Error Rate(EER).Firstly,an IPR-AHD method for iris recognition is proposed in which the overall accuracy of the algorithm is improved by using 1D Log Gabor filter together with the Adaptive subsets of Hamming distance.Previously,many researchers have used Hamming Distance(HD)approach in their iris detection algorithms.The proposed AHD as compared to the HD is more obvious and works with Hamming subsets and adaptive length.Based on density of masked bits in the Hamming subsets,each subset is able to expand and adjoin to the right or left neighboring bits.The adaptive behavior of Hamming subsets increases the accuracy of Hamming distance compu-tation and improve the performance of iris code matching.Then,1D Log-Gabor filter is used to extract and encode the data efficiently and produce a proper feature vector.For iris localization,the recognition of a circle is achieved by considering the strong edges in the image as the lo-cal patterns and finding the maximum value of a circular trough Hough transformation method.Hough transform is a standard image analysis tool to find the curves and edges which can be defined in a parametric form such as lines,polynomials and circles.Iris image is transformed into normalized iris regions by using rubber sheet model.The feature vectors obtained by using the 1D Log Gabor filter are further exploited in the recognition phase for comparison.Evalua-tion of the proposed algorithm against the state-of-the-art algorithms is performed on publicly available dataset Casia-V1.IPR-AHD shows better performance as compared to state-of-the-art algorithms.Secondly,an iris recognition algorithm based on fusion rules of FLDA and PCA is pro-posed,which merges two essential phases "localization and detection" for iris recognition sys-tems.In this algorithm,first,the image is segmented through a Hough transform algorithm to locate pupil and isolate the iris from the rest of the image.Then,the iris is normalized by using rubber sheet model and homomorphic filter is used to enhance the iris image.In this algorith-m,a fusion based on Fisher Linear Discriminate Analysis(FLDA)with embedding Principal Component Analysis(PCA)is introduced to utilize the amplitude of eye image because FLDA as compared to traditional methods can extract the features more efficiently which are used in classification process.The main contribution of the proposed model consists of enhancing the performance of segmentation and normalization processes in iris recognition systems by com-bining several descriptors of image which increases the overall accuracy as compared to the previous methods(2D-LDA+2D-PCA,2D-PCA,FLD,etc).The descriptors are complemen-tary in the sense of extracting information on the specific pattern.The extensive results show that the proposed fusion based method performs better against the state-of-the-art iris recogni-tion algorithms on publicly available dataset Casia-V1.Thirdly,iris segmentation and recognition method ISR-FMnet is proposed which is based on the architecture of Convolutional neural network(CNN).CNNs are capable of extracting and merging the non-linear low-level features of the images through the succession of layers to produce higher-level features.Neural networks implement the idea of weight sharing and reduce the number of free parameters of the architecture that directly allow fusion at different levels.Previously,many researchers have used CNN in one step of iris preprocessing(segmen-tation or feature extraction).The proposed method overcomes the existing issues in the classical methods which only use handcrafted features extraction,by performing features extraction and classification together.ISR-FMnet is based on Fully Convolution Network(FCN)and Multi-Scale Convolution Neural Network(MCNN)which merge the both steps by utilizing FCN for segmentation and MCNN for feature extraction.This architecture minimizes complexity and further improves the accuracy by adding coordinate convolutional layer and features such as Multi-Scaling and Loss function.The preliminary classification results for all categories are obtained in the fully connected layer,so that the classification results can be reflected more ac-curately.Compared with the traditional iris recognition algorithms,the proposed method has advantages and practical significance in case of training,recognition and security.Evaluation of experimental results is performed on the publicly and privatelyavailable three different datasets(Casia-Iris-Thousand,UBIRIS-V2 and LG2200).Experimental results show the superiority of ISR-FMnet over existing state-of-the-art iris detection algorithms.