High-Resolution Fingerprint Pore Extraction Method Based On Fully Convolution

At present,the fingerprint recognition technology mainly extracts the minutiae feature on the fingerprint for personal identity matching and verification.However,minutiae-based matching has some limitations and security issues.In the high-resolution fingerprint image,the sweat pore can greatly improve accurate and safety about fingerprint recognition because of its unique feature and rich quantity.In the existing research on fingerprint pore extraction,the advanced traditional methods based on sliding filtering implement multi-scale,adaptive and dynamic multi-model extraction strategies,but these methods still require complex pre-processing and post-processing.And their accuracy is also unsatisfactory,especially when dealing with poor quality fingerprint images.The methods based on convolutional neural network can better deal with the problem of polymorphic dense pore recognition because of its powerful feature expression ability.So it also can avoid the overhead of traditional method manual design features.In summary,this topic has conducted extensive and deep researches on the problem of pore extraction on high-resolution fingerprint images.The main work are following:A pore extraction method based on the object detection framework has been proposed in this paper.Two types of object detection have been used to extract pore:including the two-stage Faster R-CNN framework and the one-stage SSD framework.The experiment used appropriate strategies,such as data augmentation for dense pore,low-quality fingerprint generation and difficult sample mining.Due to using of unique strategies,the accuracy of pore extraction under the object detection frameworks is significantly improved and the effectiveness of these methods are verified.The pore extraction method based on a fully convolutional network(Pore Det)constructs a finer and more stable pore detector with multiple shortcut structures,which introduces context information around the pore and combines Focal Loss.The edge blurring strategy is adopted to realize the dense pore detection of high-resolution fingerprint images of any size.Moreover,a large number of experiments are carried out on the two-scale experimental database and evaluation criteria,which is established in this paper.The experiment proves the superiority of the method under the same conditions.The fingerprint enhancement method based on the fully convolutional network is designed to overcome the image quality problem,which occurs in the process of fingerprint acquisition,and the low-resolution fingerprint.The low-quality fingerprint problem has become one of the bottlenecks in fingerprint recognition and the biggestpain point of the current fingerprint recognition technology.In order to improve the quality of fingerprint images,this paper focuses on fingerprint quality problems such as partial blur or deformation image,pore-like noise image.And a pore-based fingerprint enhancement method based on fully convolutional network(PFE-Net)is proposed,which designs an accumulated residual learning network,combines with Laplacian loss and adopts a "shrink-enlarge-adjust" strategy to help improve the quality of fingerprint images.