| As important contents of computer vision and pattern recognition, biometric identification is widely used in government management, bank’s security, social welfare and guarantee, e-commerce, national security and other important areas.At present, many biometric fusion have become the biometric identification technology which is an important and effective solution. In this paper, face and ear recognition departure, the first study under a single biometric identification results, and then integration of two biometric identification experiments.Feature extraction is one of the key in biometric identification. In order improve the recognition rate of biometric more effectively, This paper have deeply researched the local binary pattern (LBP) algorithm for image texture, based on reading a larg number of doucments. Fatherly, this paper uses an algorithm that combination of wavelet and layered LBP operator to get the facial feature. First, this algorithm using wavelet to face images for multi-scale decomposition and the two low-frequency approximation images are combined into an integrated multi-scale image. Then using layered LBP algorithm to calculating the integrated multi-scale image, and obtain characteristic spectrum of LBP characteristic spectrum histogram statistics. Finally, the histogram is used as a facial feature for recognition. The algorithm at ORL and Yale face database has achieved100%and98.1%recognition rate, achieved the recognition rate of98.04%in Spain ear Libraries. Experiments prove that this composite algorithm has a higher recognition rate than other traditional algorithms, and has high robust in expressions, scale and illumination variations. Due to the high recognition rate, this complex algorithm find its widely application in computer vision and identity recognition.Classifier (algorithm) is one of the most important factors in biological recognition. Different classifier algorithms have great influence on the classification accuracy. In this paper, nearest neighbor and support vector machine (SVM) classifier are both used in face recognition, ear recognition, and multi-biometric recognition to compare classification accuracy. The experimental results show that the nearest neighbor method is better than the SVM when dealing with small sample data. However, SVM has the advantage of fast operation in the processing of dealing with large samples. When the training samples are large, SVM can calculate a more accurate classification model which can achieve higher recognition rate than the nearest neighbor.Besides, this paper researched the multi-biometric of face and ear and focused on analyzing the recognition accuracy of the training samples under different biometric. In order to test recognition effect in block case, the proposed algorithm using the AR face database for the two block experiments, results show that the algorithm to extract the facial features are obscured in the case is still able to provide better recognition discrimination. Finally, higher recognition accuracy was acquired from the multi-biometric experiments than single biometric experiments. Experiments of this paper not only verify the effectiveness of the algorithm but also show the fusion method is simple and feasible.Use this algorithm to deal with six internationally renowned database (3035images face the human ear), carried out thousands of different parameters of the experimental study and comparison, to obtain a high recognition rate and satisfactory identification results. Different parameters and algorithms, and validation of a multi-level experimental LBP features, the wavelet multi-level LBP characteristics, different wavelets and different classifiers in the face and ear recognition, face recognition of the role of the human ear and the effect of integration, access to characteristics and the results compared. This paper studies the follow-up study provides a rich, detailed experimental data and experimental curves. Research work with the relevant guidance and reference value. |
PDF Full Text Request