Improved Copy Move Image Forgery Detection Techniques

Abstract

In this revolutionized world where images are being used as proofs, image authentication has become a challenge today. Easily available image editing tools and softwares have made it easy for people to forge an image. Some forgery techniques fail to detect small and overlapped forged region while other techniques are not able to accurately detect forgery if the forged part has undergone geometric transformations. To overcome these issues, this paper describes two proposed method to detect forgery for small, overlapped and multiple forged regions that has undergone geometric transformation. The duplicate detection approach and the robust detection method are combined in the first proposed copy-move detection. The features of each image block can be obtained differently using the two methods. The PCA is used as the image block features in the duplicate detection approach. The robust detection technique compares pixel values to determine the features in the second way. These qualities and attributes are kept in one container. A lexicographical sort is then used to order the container. The image block sets are then filtered to eliminate any pairs that don’t reach a predetermined threshold. The remaining pair sets of an image block’s coordinates are then used to construct an image of the detection result. The technique stands invariant to post region duplication process. The technique detects multiple and overlapped copy-move forgery. Second proposed method uses SIFT to detect keypoint features. DBSCAN clustering algorithm is then applied to cluster the matched groups. Afterwards, morphological operation is implemented after outlier removal process by median filter. Finally, forged regions are localized using Linear Spectral Clustering (LSC). Hence, the technique accurately detects multiple forged region with high efficiency and is invariant to scaling and rotation.