High-resolution Fingerprint Image Registration

With the development of science and technology,people put forward higher requirements in the field of personal identification.As the highest level of security technology,biometrics has received more and more attention.Fingerprints are widely used in identification because of the advantages of stability,uniqueness and convenience.At the same time,with the increasing performance of fingerprint collection devices,sensors can obtain higher resolution images and richer fingerprint features,and the accuracy and robustness of fingerprint identification technologies are also increasing.The research of high-resolution fingerprint image has become the hot spot.Fingerprint stitching is required in the embedded system,and the fingerprint registration technique is the key to stitching.High-resolution fingerprint images bring high-precision,high anti-counterfeiting ability but also derived some other problems.For the same size of fingerprint area,the registration of high-resolution image required more computation.However,the recognition of real-time identification is very demanding.Although many research achievements have come out of fingerprint recognition,it is still difficult to meet the needs of some applications for time performance.It is of great practical value to efficiently implement fingerprint image registration.It is important to use limited resources for rapid registration.We focus on image registration,and study the method of image registration based on phase correlation.The work in this paper can be summarized as follows:1.In fingerprint image preprocessing,a faster median filtering method is proposed based on the characteristics of the noise distribution of the original fingerprint image.Then we analyze its speed advantage and prove the feasibility of this method in the fingerprint database provided in this paper.The comparison with the traditional median filtering method shows that this method can greatly improve the operation efficiency under the condition of little impact on the result.2.In the aspect of registration of fingerprint image translation,by analyzing the registration method based on Fourier transform,we find that the main influential factors of the registration are the inverse transform part.After theoretical deduction,we propose a model which can replace the inverse transform under the ideal situation,and verified the feasibility of the model.By optimizing the search strategy and parameters of the model,the registration speed is improved with little difference from the traditional method.3.We attempt to use the convolutional neural network to fingerprint image registration,find the amount of translation between the image and the rotation angle.In this paper,we construct two convolution neural networks with different depths.Each network compares the two models respectively to verify their feasibility in fingerprint image registration.