The Research And Implementation Of Recognition-oriented Iris Image Enhancement System

With the rapid development of artificial intelligence,biometrics technology continues to mature,the application scene is more and more diverse and rich.At present iris recognition is a safe and reliable biometric technology which is commonly used.But in a variety of application scenarios,due to the limitations of equipment and environment,iris images collected are often difficult to achieve the optimal image quality for recognition.In view of this problem and the actual iris recognition process and equipment,this paper is based on the existing image enhancement methods and iris recognition technology research and propose a new effective iris image enhancement system design scheme.Firstly,in order to correctly evaluate the quality of the iris image,as the basis for the enhancement of the iris image,this paper reasonably divides the quality level and proposes the classification model using convolutional neural network.Then,this paper proposes a new iris image enhancement architecture based on GAN network,using adversarial triples to add recognition constraints to generated images.Considering that iris recognition requires high accuracy and fast response speed,this paper proposes an iris feature extraction network based on depthwise separable convolutions as the recognition verification module of this system.Finally,the iris recognition application software is designed and implemented on the embedded iris recognition device,which improves the system process.The Experimental results show that the iris feature extraction network proposed in this paper has high recognition accuracy,and the speed is greatly improved compared to ordinary deep neural networks such as VGGNet.The method achieves model acceleration.After the low-resolution and defocused iris images are restored and reconstructed by the iris image enhancement architecture proposed in this paper,the recognition performance is effectively improved.