Image Tampering Detection Techniques: A Statistical Review

Author Name(s): Sonal Patil, Dr. K. N. Jariwala
Author Email: sonalpatil3@gmail.com

Abstract

Copyrights and digital signatures have boosted a wide variety of research areas in image security domain. Even though these systems are very secure, but there exists a gap between copyrighting an image, and an attacker actually using the image for their own purposes. Usually attackers will try to remove or alter with the original signatures present in the image, and thereby there is a need of forgery detection algorithms in order to track the validity of the image under test. In this paper we provide a statistical review of various algorithms used to detect forgery of images, and focus on algorithms which have good accuracy of forgery detection upon comparison.

Introduction

Image forgery detection has been a topic of research for more than 2 decades now, it includes multiple levels of processing for the input image. The info pictures can be forged pictures, non-forged pictures, the stream dependably stays steady. The pictures are gathered and named by the classes required at the yield. For instance, for forgery discovery, we require the yield to contain classes like typical, grafting forgery, duplicate move forgery among others, so we gather pictures and name them with the given classes, this progression is basic, and characterizes the precision of the general forgery recognition process, a completely chose dataset guarantees better characterization results. The gathered pictures are then given to a pre-handling and commotion evacuation square, where the pictures are cleaned of any clamors and are prepared so they are prepared for highlight extraction. The handling incorporates, picture combination, division of the picture, and any morphological structure tasks on the picture, among different advances which are typically dataset subordinate. Some datasets don’t require clamor evacuation, while others do, along these lines this progression is generally chosen according to the application originator’s need. The prepared picture is then given to the element extraction unit, where the highlights of the picture are assessed. Highlight assessment is another exceptionally basic advance, it characterizes the exactness with which highlights are assessed for the picture, numerous strategies including Speed up Robust Features (SuRF) and others have been proposed explicitly for forgery recognition so as to have better element extraction ability. Highlight assessment is typically gone with highlight determination for expansive datasets, so as to expel any repetition from the separated highlights. After component extraction, the forgery discovery classifier is prepared with the information highlights, preparing is finished with the pictures removed from the preparation set, while the genuine forgery recognition is done from the assessment square, where the prepared classifier is utilized with the information highlights from the given picture.

Conclusion

In this paper different methods of image forgery detection have been mentioned and discussed. These methods have the capability to identify forged images. In any case, a few algorithms are not viable regarding identifying actual forged region. Moreover wavelet and SuRF based methods are the most used methods for feature extraction in the forgery detection space and thus can be used in order to improve the accuracy of detection when combined with machine learning and artificial intelligence, thus researchers can focus on that area of work in order to improve the performance of the system.

907 total views, no views today

Download PDF File

About the author: admin