A cohesive forgery detection for splicing and copy-paste in digital images

A cohesive forgery detection for splicing and copy-paste in digital images

연구 분야: Analysis

학회: Multimedia Tools and Applications

Splicing and copy-paste are popular means of blind digital image manipulation. In this article, a novel identification of composite splicing and copy-paste manipulation is achieved concurrently on the forgery detection standard datasets Extended IMD2020, CASIA v1.0, and CASIA v2.0. An image under supervision is taken first, and texture-based Orientation Invariant Local Binary Pattern (OILBP) features are extricated using the Discrete Cosine Transform. The proposed technique uses an SVM classifier to decide whether the input image is spliced. Also, the proposed algorithm can check for copy-paste forgery in the image when not spliced. For copy-paste detection, Accelerated-KAZE (AKAZE) features are used to locate the replicated regions in the image. There is a copy-move forgery in the image to be discovered when the features match after post-processing filtering. Otherwise, the image is authentic. Experimental results illustrate that the performance of the proposed approach is improved than previous works. One of the significant advantages is that two types of forgeries can be detected simultaneously using the proposed cohesive approach.