Spoof detection for fake biometric images using feature based techniques

Ensuring the actual presence of a genuine legitimate trait as opposed to a fake self-manufactured synthetic or reconstructed sample is a major problem in biometric authentication, requiring the development of new and effective protection measures. In this paper, we are presenting a fraud detection m

Project Title

Spoof detection for fake biometric images using feature based techniques

Project Area of Specialization

Artificial Intelligence

Project Summary

Ensuring the actual presence of a genuine legitimate trait as opposed to a fake self-manufactured synthetic or reconstructed sample is a major problem in biometric authentication, requiring the development of new and effective protection measures. In this paper, we are presenting a fraud detection method that can be used in multiple biometric systems to detect various types of fraudulent attempts at entry. The proposed system's objective is to improve the reliability of biometric recognition systems through the use of image quality evaluation. The proposed technique uses general image quality features derived from a single image to distinguish between legitimate and impostor samples, making it optimal for real-time applications with a very low degree of complexity. In the proposed method we are using publically available data sets of iris which makes it highly competitive.

Project Objectives

Ensuring the actual presence of a genuine legitimate trait as opposed to a fake self-manufactured synthetic or reconstructed sample is a major problem in biometric authentication, requiring the development of new and effective protection measures. In this paper, we are presenting a fraud detection method that can be used in multiple biometric systems to detect various types of fraudulent attempts at entry. The proposed system's objective is to improve the reliability of biometric recognition systems through the use of image quality evaluation. The proposed technique uses general image quality features derived from a single image to distinguish between legitimate and impostor samples, making it optimal for real-time applications with a very low degree of complexity. In the proposed method we are using publically available data sets of iris which makes it highly competitive.

The extending excitement for the appraisal of biometric structures security has provoked the creation of different and various exercises focused on this critical field of research. Among the different perils analyzed, the direct and parodying ambushes have prodded the biometric system to consider the vulnerabilities against this sort of phony activity in modalities, for instance, the iris, the one of a kind imprint, the face, the imprint, and multimodal approaches. The intruder uses a type of misleadingly conveyed relic or endeavors to duplicate the direction of the bona fide customer, to erroneously get to the biometric system. The standard propelled security instruments are not amazing. These vulnerabilities have obviously shown the need to propose and make unequivocal affirmation systems against this hazard. In the present work, we propose novel programming based multi-biometric and multi-attack protection system that targets to vanquish some segment of these obstructions utilizing picture quality assessment (IQA). It isn't simply fit for working with a for the most part fantastic introduction under different biometric structures (multi-biometric) and for arranged mimicking circumstances, yet it furthermore gives a marvelous level of protection against certain non-ridiculing attacks (multi-ambush). What's more, being customizing based, it presents the standard central purposes of this sort of approach: snappy, as it simply needs one picture to distinguish whether it is genuine or counterfeit.

The major objectives of this project are to overcome biometric vulnerabilities and to make encryption secure using feature-based techniques. This project proposes a method to distinguish between real and fake images based on maximum accuracy

Project Implementation Method

In the proposed methodology the image is captured and pre-processed after which features are extracted using feature descriptors and then selected through either Particle Swarm Optimization (PSO) or the Genetic Algorithm (GA). The classifier then characterizes the image as real or fake.

Some feature descriptors utilized are as follows:

Histogram of Oriented Gradients (HOG): An element descriptor is a portrayal of a picture or a picture fix that improves the picture by separating valuable data and discarding incidental data. Regularly, an element descriptor changes over a picture of size width x stature x 3 to a component vector/cluster of length n. On account of the HOG include descriptor, the info picture is of size 64 x 128 x 3 and the yield highlight vector is of length 3780. HOG descriptors can be determined for different sizes. In the HOG highlight descriptor, the dispersion of bearings of slopes is utilized as highlights. Angles of a picture are helpful in light of the fact that the size of slopes is enormous around edges and corners and we realize that edges and corners pack in significantly more data about item shape than level districts.

Local Binary Pattern: (LBP) is an essential yet very compelling surface head which denotes the pixels of an image by thresholding the region of each pixel and considers the result a parallel number. As a result of its discriminative power and computational straightforwardness, the LBP surface manager has become a standard methodology in various applications. It might be seen as a uniting approach to manage the for the most part disparate quantifiable and fundamental models of surface assessment. Perhaps the biggest property of the LBP manager in authentic applications is its life to monotonic dull scale changes caused, for example, by light assortments. Another critical property is its computational straightforwardness, which makes it possible to separate pictures in testing ceaseless settings.

• Fast Retina Keypoint (FREAK): FREAK uses Adaptive and Generic Accelerated Segment Test key-point detectors, to classify a point AGAST builds decision trees, which are dynamically adapted based on image selection. Best pairs are found from training data with a greedy algorithm to find uncorrelated tests with high variance. Descriptors are normalized with respect to the dominant orientation

Benefits of the Project

Impact on the Environment

For any project or work being done, it is necessary to check if the environment around us is being compromised. The proposed project is environment-friendly and does not affect the surroundings of any harmful aspect. As this project is entirely based on software our surroundings are not harmed.

Impact on Society

Every project or research being carried out affects the society either positively or negatively. The proposed project is being carried out for the enhancement in security and overcoming all vulnerabilities that can cause threats. Thus this project will have a positive effect on society as people will be sure about the security of their personal information. Moreover, this proposed project can ensure security in high alert areas where highly confidential data may reside.

Based on the above speculations we can confirm that the project we have been working on has positive effects on both the environment and society.

Technical Details of Final Deliverable

We plan to submit the detailed output proofs of the accuracies found and the methods that can be used to improve the security in our detailed thesis. We will provide a detailed procedure of how it will be done and what were the drawbacks in the previous methods used.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

Security

Other Industries

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

Industry, Innovation and Infrastructure, Sustainable Cities and Communities, Peace and Justice Strong Institutions

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com