Copy Move Image Forgery Detection Using SIFT Features

Abstract

With the range of photo manipulation tools now available nearly anyone can modify an image change its interpretation by vast degree. A simple intensity and brightness adjustment can change the inference of image taken during morning or evening. An image can be called as a chronicle of visual perception. Copying and pasting a part of an image on to other part in the same image is the main essence of a copy- move forgery (CMF) and can be employed for many malicious purposes. Malicious image manipulations are harmful as they can lead to serious changes to the information that is being perceived by the human mind. Emphasis on the need for authentication of image content has increased since images have been inferred to have some cognitive e ects on human brain coupled along with the pervasiveness of images. General form of malicious image manipulations is CMF in which a region is cloned from source location and pasted onto the same image at a target location. Techniques often used to hide or increase presence of an object in the image. This need to establish detection of image originality and authentication without using any prior details from the image has increased by many folds. Many techniques to detect CMF using feature descriptors have been used in the past. Scale Invariant Feature Transform (SIFT) features are considered to be robust against Scale, Rotation, Translation (RST) and a ne transformations. We have used the approach of clustering similar SIFT feature descriptors and propose to extend the copy move region detection by introduction of segmentation mechanism for precise detection of the forged region. The proposed method for SIFT based CMF detection uses a combination of GMM and the G2NN methods to accurately detect the similar features. Once these features are detected the segmentation of the suspect region is performed which can be used for future analysis like object recognition.