Palmprint Recognition Method Based On Feature Fusion Research

Biometrics identification technology as the important way to protect the information security has aroused wide concern in the society of informatization, networking and digitization. The palmprint identification technology, with its many merits including rich information of palmprint image, simple acquisition equipment and high acceptance and so on, has been a high-profile biometrics. It has been intensively studied and widely applied in areas of police system, access control, medical research and public welfare.A number of research on palmprint recognition have been done by many scholars, and plenty of algorithms have been proposed. Key and difficult points of these algorithms are how to get high identification accuracy. However, the single recognition approach usually is hard to assure precision and robustness. So, we need to explore more effective palmprint recognition methods to structure practical and precise large palmprint identification system. Based on this, this paper presents a palmprint recognition method based on data fusion to overcome the disadvantages of the single method. Main works as follows:It summarizes common three methods of palmprint image preprocessing and their merits and demerits. Then it use an algorithm developed from preprocessing methods based on morphological operators and contour feature points. The experimental results demonstrate that the developed algorithm can accurately finish orientation and segmentation of palmprint region of interest. Palmprint image preprocessing provides the foundation for further palmprint feature extraction and feature matching.It proposes a palmprint recognition method based on LBP and log-Gabor data fusion. Firstly, it analyses the theory of LBP operators and log-Gabor filter. Secondly, it achieves palmprint recognition using separately LBP algorithm and2D log-Gabor transform. For LBP algorithm, different forms of operator are used to extract palmprint multilevel LBP histogram features, and χ2distances between palmprint samples features are computed. Then by means of nearest neighbour classifier it completes feature matching. On2D log-Gabor transform algorithm, filters of different directions are constructed. After partitioning, filting and coding of palmprint images, phase features are extracted, and Hamming distances between palmprint samples features are computed to accomplish recognition. The experiments prove the effectiveness of above palmprint texture analysis methods. At last, we fulfil palmprint recognition through LBP and log-Gabor data fusion. On the basis of extracting LBP features and log-Gabor features, it finishes the matching degree value fusion of two features in the score level by choosing appropriate weight, then uses K-nearest neighbour classifier to match and recognize palmprint. Simulation results of MATLAB format show that data fusion method makes the accuracy rate improved obviously.