Research On Liveness Detection Technology In Iris Recognition

Iris recognition is one of the most reliable and accurate biometric recognition techniques.Due to the complex external factors,iris recognition technology is widely used,but it also exposes the vulnerability of the fight against spoofing attacks.The main point of this thesis is liveness detection in iris recognition,which means distinguish imposters who use recaptured samples,wear color contact lens,or hold eyeball models.In this thesis,a near-infrared iris image dataset is established for the research of liveness detection algorithm.The main innovations are as follows:(1)For each of the three attacks,a liveness detection algorithm is proposed,and an iris liveness detection system is designed based on a cascade of traditional algorithms.Firstly,the difference between the spoofing sample image and the living iris image is used as the breakthrough point to investigate the recaptured attack,color contact lens attack and eyeball model attack.In order to defend against the recaptured attack,a recaptured iris image detection algorithm based on texture analysis according to the texture details is proposed.However,the robustness of the algorithm is not satisfactory enough.Thus,a recaptured iris image detection algorithm based on near-infrared imaging model is proposed by establishing the model of the human eye imaging under near-infrared point light source.In order to defend against the color contact lens attack,the structure and wearing characteristics of the glasses are analyzed.Then a color contact lens attack detection algorithm based on biological texture analysis and brightness change is proposed afterwards.In order to defend against the eyeball model attack,the features of spoofing iris texture and eye imaging are analyzed.Then an eyeball model detection algorithm based on structural similarity and texture complexity is proposed afterwards.Finally,a cascade classifier is built,whose F1-score reaches 0.952 as shown by the experiments.(2)A unified iris liveness detection algorithm based on the compound attention mechanism ensemble learning network is proposed.Iris liveness detection based on traditional algorithms has some problems,such as algorithm isolation,update difficulty and so on.To overcome this shortcoming,this thesis proposes a compound attention mechanism ensemble learning network(CAMEL-Net).The dual attention model is added to the backbone network to make the network pay more attention to the iris region,and an ensemble fusion block(EFB)is proposed in the two network fusion stage.The rigid fusion of feature maps will make it difficult to improve network performance,while EFB can make up for this shortcoming.The F1-score reaches 0.972 as shown by the experiments.The two kinds of iris liveness detection algorithms proposed in this thesis can effectively detect the liveness when the image is slightly defocused or the recaptured medium and the means change,so as to reduce the threat of spoofing samples to the iris recognition system.