Research On Iris Recognition Method Based On Wavelet Packet And Neural Network

Biological characteristics are unique,safe and easy to carry,and they have been widely used in our daily life.Different biological characteristics have their own characteristics in different aspects.Because the iris has a very complicated physiological structure,everyone's iris texture characteristics are different,so the iris is safer,more stable than other biological characteristics,and it is very difficult to forge.At present,there is still much room for improvement in iris recognition technology.Therefore,this paper proposes an iris recognition algorithm based on the combination of wavelet packet transform and neural network for the unique physiological structure of iris.The iris image collected by the device includes not only the iris area,but also the eyelids,eyelashes,sclera,and pupils.During the iris recognition process,the iris area must be segmented to eliminate these useless noises.This paper improved U-Net network to segment the iris.Add a hole convolution with different hole rates to the dense block structure,and add it to the U-Net as a dilated convolution dense block structure to make full use of the detailed information of the image at different scales,so that the feature map in the network can be better connected.The iris segmentation experiment was carried out in the two data sets of CASIA-4i and UBIRIS.v2.The error rate of e1,e2 and the accuracy of f1 were used as the evaluation indexes of the segmentation results.The experimental results proved that the network proposed has excellent performance and it has better accuracy and robustness.After the iris image has gone through a series of preprocessing operations,this article first performs a block operation on the iris image,and then performs a two-layer wavelet packet decomposition operation on the iris block image.According to the vertical distribution of the normalized iris texture,this paper chooses to extract features from the four sub-band images of the horizontal low frequency and vertical low frequency,intermediate frequency and high frequency after wavelet packet decomposition,and then use their singular values to construct the features vector,Euclidean distance is used as a feature matching method to verify the effectiveness of feature selection.Finally a neural network is introduced as a classifier,and the singular values vector is used as the input of the network for classification.In the recognition experiment,the correct recognition rate is used as an evaluation index to discuss the influence of different block methods on the correct recognition rate.According to the recognition results,the different singular value vectors selected for each sub-band image are compressed,and the appropriateness hidden layer of neural network is finally determined.Finally,a better correct recognition rate was obtained,and completed the iris recognition algorithm.The iris recognition algorithm in this paper has a higher correct recognition rate compared with other algorithms.