Biometrics Methods Based On Palmprint And Hand Shape

With the development of economy, demands for biometrics techonology areexpanding, and systems with high reliability and high acceptability are current focus.Biometrics based on hand features is highly reliable and highly acceptable, and thefusion of palmprint and hand shape could result in a more reliable system.Consequently, in this paper technologies of palmprint recognition and hand shaperecoginiton are investigated independentlly, and the common fusion methods areextensively evaluated. Specifically speaking, this paper includes works as follows:(1) For palmprint recognition on original images, a maximum margin criterionmethod with tensor representation is proposed. As one branch of subspace-basedmethods, this method completes the family of maximum margin criterion of differentrepresentations. This method has competent recogonition performance as othertensor-based methods but is more robust. Compared with maximum margin criterionmethod with vector representation, the proposed method is much more efficient. Theproposed method could achieve high recognition performance with few trainingsamples, which is very suitable for biometric applications. The effetiveness of theproposed method is evaluated on palmprint database.(2) For palmprint recognition on feature images, two feature representations basedon histogram of oriented line are proposed, which are based on Gabor feature andfinite Radon transform, respectively. Applying subspace analysis methods on theserepresentations could achieve better recognition performances than ones on originalimages. The proposed two representations make the subspace-based methods achievea similar recognition performance as top-level palmprint recognition methods, whichare highly reliable and applicable.(3) For hand shape recognition, an improved method based on shape contexts isproposed, where a novel coherent distance is introduced specifically for the handshape recognition problem. The proposed method is robust for different position and placement of the fingers, and could acheve the highest recognition performance ondatabase of the same size. The proposed method is efficient enough for real-timeprocessing in verification system, and could satisfy the demand of real-timeprocessing in identification system to some extend.(4) For the proposed palmprint recognition methods and hand shape recognitionmethods, common fusion methods on score level are extensively evaluated, and thebest fusion method is determined, which completes the proposed multimodalbiometrics system based on palmprint and hand shape.