Research On Iris Feature Extraction And Recognition Algorithm Based On Improved DenseNet Network

Biometrics feature recognition is a popular technology in the field of identity authentication and identification.Iris recognition technology stands out among many biometric recognition technologies.And it has become the focus of people's attention because of some of its huge advantages such as high security,stability,uniqueness and anticounterfeiting.Iris recognition technology includes two parts,iris preprocessing and iris feature recognition.Iris preprocessing includes iris image acquisition,iris image quality evaluation,iris localization,iris normalization and enhancement.Later part includes iris feature extraction and iris feature recognition.For the Iris feature extraction and recognition,in this paper we provide DenseIrisNet which is a network model based on deep learning,is proposed for iris feature extraction and recognition.The following is a brief introduction of the work of this paper through iris feature extraction,iris feature matching,experimental analysis and other aspects,1,At the stage of iris feature extraction,a network model structure based on DenseNet was proposed.To solve the problem that the size of the iris image feature receptive field area is limited in the model structure Dense Block module and improve the model structure,the Inception multibranch structure is added to the model structure.The structure is composed of convolution kernels of different scales,and the features of different branches are fused.At the depth of this model structure,using the idea of iris feature reuse,extract important feature information from the current layer and transfer it to the next layer of the network in turn.At the width of this model structure,compared with the conventional convolution operation,the structure is composed of multiple branches.So the iris features can be extracted through a convolution kernel of different scales to obtain richer feature texture information.Then the scale of extracting iris feature information is expanded,and the amount of model parameters and calculation cost are reduced.2,At the stage of iris recognition,the Euclidean distance matching algorithm is used for training the DenseIrisNet network model proposed in this paper on the iris database.Through the learning and training,the network model can automatically extract iris feature information and generate feature codes.The coding dimension of the feature coding is set by k which stands for the value of the growth coefficient of the network model.The feature codes extracted by the DenseIrisNet model are combined with the Euclidean distance matching algorithm to calculate the distance similarity,and the iris feature matching recognition is performed.3,In the experimental part of this paper,JLU-6.0 Iris library provided by laboratory of biometrics and information security technology of Jilin university and CASIA-Iris-Lamp Iris library provided by institute of automation of Chinese academy of sciences were used as the experimental database.In the experiments,the performance evaluation indexes of the iris recognition algorithm are based on the correct recognition rate,equal error rate,and ROC curve.The iris feature extraction performance of the DenseIrisNet network model proposed in this paper is analyzed,and the overall comprehensive performance of the iris recognition system is also analyzed.Through comparison experiments with existing traditional iris recognition algorithms,it shows that the algorithm proposed in this paper has improved the correct recognition rate of iris recognition and the performance of network models.