Research Of Non-contact Low-resolution Palmprint Recognition

In recent years,with the development of science and technology,dissemination of the information on network has become more and more convenient and the information security problems begin to trouble people simultaneously.The traditional identification technology has not been able to meet the needs of growing information security.Biometric identification is a technology that through identificate the inherent characteristics of the human body to determine the identity of the person.Biometric technology is safe,convenient and accurate and has been widely used in identity certification.Compared with other biometrics,it has the advantages of high recognition rate,fast recognition speed and low equipment cost characteristics.Therefore,it has become a hotspot in the field of biometrics.When using the palmprint recognition syste m,the palmprint images collected by the system are not exactly the same with the palmprint images collected at the time of verification.During the collection of palm prints,the plamprint images will shift and rotate if your hand placed in different positions.Brightness of images will be different caused by the different of the light intensity.The palmprint image will be deformed if the palm are not fully expanded.The existing palmprint recognition algorithm can not solve the above problems well and leading to user authentication failure.To solve the problems mentioned above,this study presents two kinds of palmprint recognition algorithms.The first palmprint recognition algorithm based on gabor filtering and directional coding and blocking and another based on MFRAT filtering and directional coding and blocking.The main works of this study are as follows:First,In this research,a half-Gabor filter with twelve directions and a half-MFRAT filter with twelve directions are used to filter the palmp rint image in order to obtain the filtered response in each of the pixels in the palmprint image.Even if the noise in one direction affects the filtering response,the responses of the other eleven directions ensure the print can still be identified.Second,twelve directions of the filter response were used to the palmprint image recognition and the twelve directions are encoded.The coding of the directions makes the palmprint recognition algorithm proposed in this study has good robustness to the rotation of small angle for palm images.The directions are encoded as zero or one,owing to the bit operation is performed at the time of matching,the speed of the algorithm is also improved.Third,as to image matching,the block matching strategy was proposed in this study.The image blocks were obtained from the point on the main line of the palmprint image,and were matched corresponding to the two images.The image blocks were translated at the time of matching and the translation parameters of all the blocks were used to calculate the final palmprint matching score,so that the algorithm is robust to the translation of the image and can improve the degree of discrimination between the true matching score and the false matching score.Finally,the recognition accuracy of the algorithm was improved.Experiments were performed on the Poly U database of the Hong Kong Polytechnic University.The EER of the classical palmprint algorithm was 0.16% and the EER of the two algorithms mentioned in this paper was 0.083% and 0.094% respectively.Experiments show that the algorithms proposed in this paper have higher recognition rate than the classical palmprint algorithm.