Finger Vein Feature Extraction And Matching Algorithm Based On Correction Of The Intensity Inhomogeneity

Biometrics integrates computer,biosensor and other technologies,and uses inherent body characteristics for authentication.Common biological characteristics include vein,fingerprint,face,pupil,etc.Compared with other biological characteristics,the finger vein characteristic information is hidden under the human skin,which is not easy to be contaminated,destroyed or stolen.Hence it is more secure and has gradually become a research hotspot in identity recognition.However,there is no standardized and unified process for collecting finger vein images,and problems such as uneven exposure and excessive noise often occur.To solve these problems,this thesis studies the image feature extraction and matching algorithms,and does the following work.(1)This thesis analyzes the feature distribution and pixel characteristics of finger vein images,introduces a fuzzy C-means clustering(FCM)algorithm,and proposes an FCM based intensity inhomogeneity correction algorithm(FCM-IIHC).Simulation results show that traditional intensity inhomogeneity correction algorithm applied to vein images may cause feature loss.The FCM-IIHC algorithm not only reduces the loss of features,but also achieves a better correction effect.The finger vein image processed by FCM-IIHC algorithm has better matching performance than the traditional method in the same image matching scenario.(2)By studying the Speeded-Up Robust Features(SURF)algorithm,this thesis proposes a SURF based feature extraction and matching algorithm(SURF-FEM).The algorithm not only combines the characteristics of the finger vein image processed by the FCM-IIHC algorithm,but also combines the characteristics of the Shi-Tomasi corner point and the M-estimators SAmple Consensus(MSAC)algorithm.Simulation results show that SIFT takes a long time,and SURF is not effective for lowquality finger vein images.The SURF-FEM algorithm proposed in this thesis solves this problem.It may achieve performance close to or better than SIFT with less time complexity,and has better robustness in offset,noise,and rotation scenarios.