Biometric Security Solutions for Human Authentication

Abstract

In the present day of automated world, machines are replacing the human in every aspect of life. Due to this, the security concern regarding the authenticity of the user goes on increasing. Hence, it becomes necessary to include some constrains in order to reject imposters (unauthorized persons) and allow only the authorized user to access automated services. Biometric can provide the solution to these problems. Although a lot of work has been done in the field of biometric, particularly, in the field of fingerprint biometric, nevertheless objective of highly secure practical solution is still to be achieved. Present thesis is an attempt to propose biometric based security solutions for human authentication with improvement in existing techniques and solutions based on level of security concerns. In the initial part the major problems in minutiae based fingerprint authentication system have been addressed. The improved algorithms for thinning, minutiae extraction and relative alignments of query and reference fingerprint images have been proposed. In addition to that an image based fingerprint verification system; a combination of biometric (voice) with conventional method (password) and a fuzzy logic based multibiometric system using fingerprint and palmprint has also been proposed. An improved thinning algorithm (preprocessing step in minutiae based system) has been proposed and implemented by using the Karnaugh map (K-Map) technique in which all the rules are simultaneously applied to each and every pixel which resulted in faster response as compared to look up table technique. The response time of the system has further been reduced by improving the window extraction time and by using the short circuit logical operators. Single isolated pixels are also removed from thinned image as they are not required in the final minutiae extraction stage. Rotation independent thinned skeleton of one pixel width and overall significant reduction of CPU time has been achieved with the proposed method. Crossing Number method is one of the most widely used binarization based method for minutiae extraction but is highly depends upon the pre-processing steps and is not robust against the spikes (a non minutiae point which is required to be eliminated in the post processing step). A new method has been proposed for the minutiae (feature) extraction. In the proposed method only the genuine cases of minutiae have been identified and solved to a minimized logical vi expression using a well known minimization technique of Karnaugh map. The capability of the proposed algorithm has further been enhanced by including the steps to remove the boundary pixels. The proposed method eliminates up to 25% false minutiae in the extraction stage, which remains with the crossing number method. In order to find the rotation and/or translation between the reference image and query image a Genetic Algorithm (GA) based relative alignment algorithm has been proposed. In the proposed algorithm there is no need to find the reference core or delta point because reliable detection of these reference points is a difficult task. In the proposed algorithm, all the three parameters x, y (translation) and  (rotational) have been optimized separately. The processing time of the GA based algorithm has been improved by considering the range of deviation of the query data point from the reference data point and by using the binary search algorithm. An image based fingerprint verification system using LabVIEW (Laboratory Virtual Instrument Engineering Workbench) has been implemented. The proposed method uses a learning phase, which is not present in conventional image-based systems. The rotation and translation between the query and reference image has been taken care by calculating and comparing the circular intensity profile of both the images. Further in order to increase the speed of the system, sub sampling of the image has been performed which reduces the amount of data required for matching. The size of the template is very critical for the success of image based system. The simulation results had been obtained for three different sizes of template images (i.e. 50 50, 100100 and 200200) and for different thresholds (threshold of 700, 750, 800, 850 and 900). For the smaller size template image (50 50) the false rejection rate is less in comparison to larger learning images but false acceptance is high. So, a compromise has to be made between false acceptance and false rejection. Simulation results for different fingerprints with various learning image sizes of FVC2002/Db1_a database reveal that a 100100 learning image size for a threshold value of 700 (1000 being the perfect match) gives good results for false acceptance rate (FRR) and false rejection rate (FAR). A FRR of 1.027% and FAR of 0% have been achieved with 100100 sized template image at a threshold of 700. Further, the fingerprint images have been enhanced using the STFT analysis and contextual filtering in the Fourier domain and processed in the similar way to find false acceptance and false rejection rates. On comparing the enhanced database results with the original database results it has been observed that the later (original) are better than the former (enhanced). This is due to the fact that vii although with enhancement ridge- valleys structure of the fingerprint improves but much richer grey-level information of a fingerprint image has been lost. Finally, in order to overcome limitations of single biometric such as non uniqueness, noise and intra class variation etc., two solutions have been proposed. In the first solution, a system integrating a single biometric (speech) with conventional method (password) has been developed. In the developed system, the first stage identifies the group of persons on the basis of biometric (speaker identification) and second stage authenticates the person on the basis of password from the list of selected entries in the first stage. The proposed integrated system has increased the accuracy of the system to 99.875% (with ten entries extracted in the first stage) and at the same time also overcomes the problem of non-enrollment of the user. In the second solution, two biometric traits, palm print and fingerprint have been combined at the score level using Fuzzy logic based system. Three different sets of if-else rule have been formulated for low, medium and high security. In order to overcome the problem of non universality the rules for the low security system, have been formulated in such a way that if any of the inputs is with a reasonable score then the system accepts the claim of the user. With this type of arrangement 0% false rejection rate and 0.75% false acceptance rate has been achieved. With the formulated rules of medium and high security false rejection rate increased to 1.61% and 2.88% respectively from the 0% for low security but false acceptance rate get reduced to 0.225% and 0% respectively. Further, a two step very high security system has also been proposed in which another biometric trait (voice) has been added with the palm and fingerprint to enhance the security. The three biometric traits have been combined in such a manner that the voice biometric identifies the person while the combination of palm and fingerprint authenticates the person. Findings of this research work can be utilized to improve existing human authentication systems.