Research On Iris Recognition Method Base On One-dimension Wavelet Feature Combined With BP Neural Network

Nowadays,with the rapid development of the information age,the identification and authentication are becoming more and more inseparable in human society,and higher requirements are placed on the convenience,accuracy,security and reliability of identity authentication.Under the demand,iris recognition technology emerged as the times require,and has been continuously developed and improved in recent years.High-efficiency recognition is the goal that the iris recognition field has been pursuing,and it is also the requirement of iris recognition in this era.In order to improve the recognition speed and ensure the correct recognition rate,this paper adopts the method of radial folding block and circumferential period block to perform one-dimensional haar wavelet transform to extract low-dimensional iris features,and combines BP neural network for pattern matching.The work of this paper mainly includes the following four aspects:(1)In the aspect of feature extraction,the traditional method which is adopted to directly transform the enhanced image of the iris is abandoned.Instead,the normalized iris is firstly divided into a radial fold and a circumferential cycle,and the transformation is performed.The purpose of this is to reduce the iris feature dimension while reducing the sensitivity of the iris region to noise and to ensure that the iris effective features are not neutralized.(2)One-dimensional haar wavelet transform for periodic extension of the image after segmentation is used to extract high-frequency information.The purpose of periodic extension is to overcome the sensitivity to rotation in iris recognition,and the high-frequency part after wavelet transform can well preserve the details of the signal.(3)In order to verify the effectiveness of the block method and wavelet decomposition feature proposed in this paper,the traditional wavelet zero-crossing detection is used for feature coding.The Hamming distance is used to calculate the similarity,and the statistics are divided into blocks and non-blocks and different blocks.The discriminability value of the block and the number of different wavelet decomposition layers is used as an evaluation index.Finally,the correct recognition rate,error rejection rate,equal error rate and distinguishability value are calculated when the number of layers is optimally divided and decomposed.(4)In order to further improve the correct rate of iris recognition,this paper introduces BP(Back Propagation)neural network classification method in the pattern matching stage.Using the BP neural network algorithm in machine learning to learn the iris library,using the obtained classifier to classify,can avoid the disadvantage that all feature points have the same influence on the recognition results.This paper first sets the hyperparameter of BP neural network to the iris.The high-frequency information of the haar wavelet transform of the library training set is directly trained as the input of the BP neural network to obtain the classifier,and the final correct recognition rate is obtained by using the test set.The experimental data in this paper,including the database used in the verification feature validity stage and the verification BP neural network classification effect stage,is the fourth generation iris library of Jilin University independently researched and developed by the laboratory: JLU-4.0;in order to ensure the reliability of the comparison results,the Iris-Interval sub-library of CASIA-IrisV-3.0 and the JLU-4.0 iris database were used in comparison with other algorithms.In summary,this paper proposes a block-based preprocessing method,which extracts features by periodic extension wavelet transform,and finally solves the multiclassification problem by using BP neural network,reduces the feature dimension,and improves the correct recognition rate.