A Note on Authentication Accuracy with Multiplicity of Biometric Images

This project cum paper we have implemented basically deals with authentication and the various improvements that can be brought about by integrating them with real time scenarios. Our basic aim of it to implement is to develop a secure environment that cannot be breached easily. There are many applications or algorithms that provide authentication at the finest level but we are a step ahead I believe because of the uniqueness of the idea and effort put behind it. Let me give you a brisk knowledge of what we have done with it “by increasing the number of biometric images for authentication, authentication accuracy is expected to be improved”. This paper considers simple algorithms for verification and identification with multiple biometric images for each person. INTRODUCTION Personal authentication has been an essential issue in many social infrastructure systems. Biometric authentication has attracted attention as a technology to compensate some weaknesses of tokenand knowledge-based authentication. With the spread of computers and networks, the scope of applications of personal authentication was extended into a wide area, and the number of persons who use each application system is supposed to become huge. Especially for biometric authentication, accuracy of personal authentication becomes an important factor by the increase of the number of persons. The aim of our research is to find a way to improve accuracy of biometric authentication. One of the straightforward approaches for the improvement is increasing the amount of information for authentication such as the number of biometric images registered in an authentication system (such registered images are called “templates”). This trial to improve authentication accuracy by several biometric images can be a typical application of statistical analyses. However, in order to apply such analyses appropriately into biometric images, some knowledge of biology about the part of human beings or of image processing about the feature extraction will be required after all. We simply focus on the results of comparisons of biometric images. Our approach is on the assumption that the basic authentication with a single template for each person is conducted on the basis of the result of a comparison between the template and an inputted image, and then the effects of the number of templates on accuracy are evaluated as the change from the basic authentication. Then, we consider two simple ideas in order to treat the results of comparisons for multiple templates, that is, a majority vote and the arithmetic mean of the similarities. There exist two possible procedures of biometric authentication, that is, verification and identification. At least for identification, the effects of the number of templates for each person on accuracy are not trivial even for the simple methods to treat multiple results of comparisons. In this project cum paper, we define simple algorithms for verification and identification based on the idea of a majority vote and the mean of similarities. And then, we apply the algorithms to practical palm print images in order to examine the error rates as accuracy of authentication. In order to measure the similarity of two images, we consider the matching of the features extracted by Scale-Invariant Feature Transform (SIFT). There already exist some researches that apply SIFT to authentication with biometric images such as fingerprints and palm prints. Additionally, it is expected to be applicable to general comparison based authentication algorithms with multiple biometric images. WORKING This section defines working of our proposed system and how algorithms are devised for image matching which being an essential part our research. PROFILE VECTORS The proposed system creates user profile as followsMaster vector – (User ID, Sound Signature, and Tolerance) Detailed Vector – (Image, Click Points) As an example of vectors Master Vector (Varun, Dolphine.wav, 10) Detailed Vector Image Click points I1 (123,678) I2 (176,134) I3 (450,297)