Research On Iris Localization Algorithm Based On Facial Landmarks Detection

Iris recognition plays a vital role in today's personal identity authentication work,and iris localization is a key part of iris recognition technology.The process of iris recognition depends on the inner and outer boundaries of the iris selected during the positioning stage.Excellent iris localization algorithm can effectively improve the accuracy of the recognition stage and is of great value to the iris recognition technology.The topic of this paper comes from the iris localization problem in the actual scene.It aims to solve the iris localization problem of the face in the actual scene.Factors such as lighting,occlusion,and posture in actual scenes will challenge the robustness of the iris localization algorithm,making the application of the positioning algorithm in reality impede.In order to solve the above problems,this paper proposes an iris localization algorithm based on facial landmarks detection,which innovatively uses the information of face landmarks to assist the iris localization,effectively overcoming factors such as threshold setting and noise interference.The impact of performance,the algorithm of this paper has a certain accuracy and robustness.The main work of this article is:First,This paper analyzes the actual engineering project,which aims to perform tasks such as face detection,iris localization,three-dimensional reconstruction of the face,and three-dimensional pose estimation of faces in pedestrian photos.This paper focuses on solving iris localization task,and designed an iris localization technology solution based on facial landmarks detection,which can output intermediate results while completing iris positioning: facial landmarks,this information can assist the project in the human face 3D pose estimation task,the design of the algorithm has important engineering significance to the overall project.Second,we collect and label facial images by ourselves,and construct a 24-points facial landmarks dataset.The labels include the coordinates of facial landmarks,facial pose angle,facial attributes,iris area,etc.On the basis of the above dataset,a facial landmarks detection algorithm based on cascade network was engineered to solve the imbalance problem of different facial features positioning difficulty in landmarks detection tasks.Perform coarse positioning and fine positioning of the eye area,and compare the experimental results with other related algorithms.Third,This paper proposes a facial landmarks detection algorithm based on facial pose and attribute labels.The algorithm uses a fully convolutional network as the main network,which improves the defects in the process of obtaining the coordinates of landmarks from landmark heatmaps,and innovatively proposes a loss function based on facial pose and attribute labels The training of the network makes the network more able to focus on difficult samples with large poses and exaggerated expressions.A large number of experiments on the dataset show that the algorithm can effectively ensure the accuracy of the model while controlling the size of the model,and has great practical application value.Fourth,this paper presents an iris localization algorithm based on the improved Daugman algorithm,analyzes the defects of the original Daugman algorithm,uses the facial landmarks as a priori information to assist iris localization,and proposes an improved iris image preprocessing algorithm to eliminate the specular reflection in the iris,which improves the accuracy of iris localization.A large number of experiments on the dataset show that the algorithm has certain robustness and general performance.