High-precision Contactless Palmprint Recognition Technology Research

In the age of informationalization,how to identify a person's identity and protect its stability and security has become an increasingly important issue.Palmprint recognition is an emerging biometric recognition technology.Compared with other identification technologies,palmprint recognition has lower requirements of the environment,it is easy to collect the palmprint with low-cost devices.However,palmprint recognition is an extremely complex process involving multiple steps.In order to solve the image alignment problem,many local invariance features are proposed.The local invariance features have the characteristics of rotation invariance and affine invariance,which can be widely used in practical fields such as object tracking and image registration.However,its complexity and large amount of calculation make it difficult to be widely used in non-contact situations systems.In the face of the massive palmprint dataset and huge application requirements,a robust algorithm with high accuracy and speed is urgently needed when designing the palmprint recognition system.Aiming at the problems existing in the current non-contact palmprint algorithms facing,this paper propose two solutions.The first still use local invariance features.The performance of different kinds of local invariance features on palmprint images are compared.The match speed of those features are very low due to their violent matching strategy.In order to further decrease the time cost of the feature extraction and matching operations.Another method uses the variant of the local invariance features and utilizes patch matching strategy in the matching stage.Both this two methods utilizes the characteristics of local invariant feature.However,the second method also achieves the characteristics of high recognition accuracy and speed of the traditional palmprint algorithm base on palmprint-line coding.This method uses the Dense SIFT feature to encode the palmprint image,and matching the encoding feature hierarchically.The hierarchical search algorithm bases on the pyramid structure of the encoding palmprint image,and combines the advantages of random search and neighbor propagation.Our approach can effectively solve the misalignment problem of the non-contact palmprint image.And it obtains EER at 1.75% on IITD database.For the palmprint image with an input size of 128×128,the Dense SIFT feature encoding takes 170 ms,and the pyramid search takes 7.1ms.This method is sufficient to meet the challenges of big data application.