Research On Bimodal Feature Fusion Method Of Finger

With the continuous development of science and technology,information security is playing an increasingly important role in people's daily life.Biometric identification technology has gradually replaced the traditional identification methods due to its convenience,reliability,and security.However,the unimodal biometric identification technology using only one biometric feature is easy to be forged and deceived in practical application,which makes it insufficient to meet the growing security needs of people.Multimodal biometric identification technology can overcome the problems existing in unimodal biometric recognition to some extent because of the integration of different biometrics.In recent years,multimodal biometric recognition technology has made rapid development,but how to achieve effective fusion between different biometrics is still one of the key problems that need to be solved in multimodal biometric recognition.Therefore,in this paper,we take finger-vein images and fingerprint images as the research objects to study an efficient and reliable method of finger bimodal feature fusion on feature level and establish a convenient and efficient bimodal biometric recognition system at the same time.The main works of this paper are as follows:(1)In order to solve the problem that the offset and rotation of the original fingervein images have great influence on the subsequent recognition results,a method to extract the region of interest in finger-vein images is proposed in this paper.The proposed method segments the finger region by dynamic threshold to reduce the influence of background noise on recognition results,and on the basis of using the least-square method to fit the finger midline to achieve finger region rotation correction,the location of the region of interest is determined through the pixel concentration distribution of the finger region,so as to improve the effectiveness of finger-vein recognition.Based on this,a unimodal biometric recognition method based on double-weighted group sparse representation classification is proposed for the problem of large intra-class differences and small inter-class differences on finger-vein images and fingerprint images.The proposed method adds the constraints to sparse coefficients in both individual and group levels to improve the recognition performance of low-quality finger-vein images and fingerprint images,and also lays a solid foundation for the subsequent research on bimodal biometric recognition.(2)In view of the lack of discriminability and intra-class variation in unimodal biometric recognition,a bimodal biometric recognition method based on weighted joint sparse representation classification is proposed in this paper.On the basis of achieving the fusion identification of finger-vein and fingerprint by sharing sparse representation,the proposed method fully considers the locality of data within each modality.In this method,a penalty function is constructed by using the Euclidean distance between each modal training sample and the test sample,and a penalty constraint is added to the joint sparse coefficient to avoid the influence of long-distance training samples on the final recognition results,so as to improve the performance of bimodal biometric recognition and the effectiveness of identity recognition.(3)In order to further improve the recognition efficiency of weighted joint sparse representation classification,a dictionary optimization method based on k-NCN is proposed in this paper.In this method,the k-NCN algorithm is used to reduce the number of training samples in each modal sub-dictionary,which reduces the time for solving the joint sparse coefficients and improves the final recognition efficiency.On this basis,for the problem that the weighted joint sparse representation classification method ignores the training sample label information and the data integrity within each modality in process of fusion recognition,a bimodal biometric recognition method based on doubleweighted joint group sparse representation classification is proposed in this paper.The proposed method not only makes full use of the label information of the training samples,but also fully considers the locality and integrity of the internal data within each modality,which effectively improves the performance of finger-vein and fingerprint fusion recognition.(4)In order to provide richer discriminative information for subsequent recognition and classification,the paper proposes a multi-feature fusion method based on canonical correlation analysis to achieve the organic combination of texture features in the spatial domain and frequency domain of each modal image.Meanwhile,to address the problem of ignoring the differences between various modal features in the k-NCN-based dictionary optimization method,we propose a dictionary optimization method based on score level fusion.The proposed method fully considers the differences between finger-vein features and fingerprint features.On this basis,in order to further improve the effectiveness of finger-vein and fingerprint fusion recognition,a bimodal biometric recognition method based on multi-weighted joint group sparse representation is proposed in this paper.The proposed method not only takes into account the internal data and weighted scores of each modality,but also makes full use of the quality scores between the images of each modality,which further improves the accuracy and reliability of bimodal biometric recognition.