The Research On Multi-model Handmetric Algorithms And System Implementation

Biometrics refers to automatic recognition of an individual based on her behavioral and/or physiological characteristics, which can be widely used in security systems for entry control, attendance recording, government affair administration and identity authentication of digital terminal access and network resources access control etc. Multimodel biometrics integrates the evidence presented by multiple kinds of biometric traits of a user, which can improve the system performance in many aspects including accuracy, noise resistance, universality, spoof attacks, and reduce performance degradation in huge database applications compared to systems based on a single biometric modality. Multimodel biometrics is a promising new field and becoming the focus of recent studies.The proposed "handmetric" includes 2D-finger-geometry, knuckleprint and palmprint. Based on the researches on unimodal biometrics respectively, in order to impove the time efficiency of recognition in large database, an AND rule decision level fusion scheme is adopted, a hierarchical matcher from coarse to fine, from fast to slow is designed, so that an efficient multimodal biometrics system is realized with high recognition accuracy and fast searching speed. Experiments were carried out and the evaluation results validated the effectiveness of the proposed system. An embedded palmprint verification system is explored and realized finally. Main content of this work is as follows:1. An on-line hand image capture system based on webcamera is designed and a robust hand image preprocess method is proposedThe schema and architecture of the hand image capture device is given. The device has advantages of simple interface with PC, low cost, friendly with user, stable to environmental changes. A robust palmprint ROI and fingers' ROI extraction method is introduced. Meanshift filter and Ostu binarization method are used in the preprocess procedure. A novel palmprint ROI extraction method based on pivots of index finger and little finger is employed, which eliminated translation and rotation variation in certain extent.2. A simple and efficient 2D finger geometry recognition algorithm is proposed2D finger contours are used as distinct individual trait to authenticate personal identity. A novel kind of finger geometry feature is constructed which has merits of distinctive, robust, low dimension, and is characterized with a high convergence of intra class similarities and good dispersion of interclass discrimination. This identity authentication method exhibits short matching time due to small feature size as well as high recognition accuracy, which is appropriate for initial match in large database to find out the most similar samples for further matching.3. Personal recognition algorithm based on knuckleprint is investigatedThe feature extraction method based on vertical projection for binarized gradient image and wavelet denoising is proposed, which decreases representation dimension from the 2D image to 1D vector. The matching procedure is performed by measuring the similarity between two feature vectors. This recognition method based on knuckleprint is small in feature size, low time cost and highly accurate in recognition, which is suitable for high level coarse match in large database to search most similar samples for further fine level matches.4. Researches on palmprint recognition based on principal lines and dual tree complex wavelet texture features are carried outFirstly, a probability distribution template of principal lines is generated using fuzzy method, and bidirectional matching distance is used to measure the similarity. Then dual tree complex wavelet is applied onto palmprint image blocks to obtain texture feature and Canberra distance is used for matching. Since principal-line-based recognition can be used to discriminate the significant palm feature and dual tree complex wavelet, with good property of approximate shift invariant and multi-directional selection characteristics, they can be used to detect palm texture efficiently. These two features complement each other very well. Experiments suggest that the palmprint recognition algorithm based on principal line feature and dual tree complex wavelet texture feature has low time complexity and good flexibility.5. Research on translation, rotation, scale invariant palmprint recognition algorithm is carried outA palmprint recognition algorithm which can recognize palms correctly from palm images with great variations in translation, rotation, scale is proposed for the first chance. SIFT operator is introduced into palmprint recognition. For the low resolution and quality images, KLT corner detector is used for keypoints extraction, directional independent normalized local descriptor is constructed for each keypoint, which is invariant to image translation, rotation, and illumination changes. Experiments proved the effectiveness of the algorithm, which can use low-cost digital camera to capture palm image, extract valid palmprint region from complicated background and realize palmprint recognition. Moreover, the algorithm is invariant to image translation, rotation, partly invariant to distance, gesture changes of the palms, further proved its robusticity.6. A multimodal handmetric decision level fusion based on hierarchical matches is explored and realizedA multimodal biometrics system scheme based on hand traits is given. The scheme integrated the high efficiency of finger geometry and knuckleprint recognitions with the high accuracy of palmprint recognition. Among the existing multimodal biometrics information fusion methods, a simple and feasible AND rule decision level fusion method is adopted. With hierarchical matches and threshold adjustment of each matcher, not only the recognition accuracy but also the large database searching efficiency of the implemented biometrics system is improved compared to unimodal ones. The multiple features extracted from the hand image can be applied for guided search in large database. Five level features include the 2D finger geometry, knuckleprint, principal lines of palmprint, dual tree complex wavelet texture feature of palmprint, and scale invariant local descriptors extracted from palmprint image, which have been introduced in previous chapters. The high level coarse matches can reduce significantly the amount of samples that enter into lower level fine matchers, the final level matcher realizes high accuracy verification.7. An embedded palmprint verification system is initially explored and researched A practical schema of palmprint identification system based on ARM and WinCE is designed and implemented. The system not only can be embedded in entry control or security system as an independent unit, but also can be used for identity authentication as a part of intelligent family applications or intelligent terminal equipments. Experiments proved that the proposed system could identify palmprints with high accuracy. Compared with the PC palmprint recognition systems, the embedded systems has the merits of low-cost, small in size, light weight, low power consumption, convenient for moving and operation, and feasible for integration. The friendly user interface of Windows CE operation system further improved the system's user acceptance.This paper proposed three types of biometric authentication technologies according to three hand biometric traits: 2D finger geometry, knuckleprint and palmprint. Among them, three feature representations are introduced for palmprint. These five modals of handmetric fused in decision level, to improve performance of the searching efficiency in large database as well as the recognition accuracy through optimized combination. An initial attempt in embedded palmprint verification system is carried out. The experimental results demonstrate the power of handmetric recognition and proposed fusion method, and show the feasibility and effectiveness of the system. The work can put forward a new field for further study of multimodal handmetric recognition system.