Adaptive partitioning‐based copy‐move image forgery detection using optimal enabled deep neuro‐fuzzy network. (17th November 2021)

• Record Type: Journal Article
• Authors: Mariappan, Geetha; Satish, Aravapalli Rama; Reddy, P. V. Bhaskar; Maram, Balajee
• Is Part Of: Computational intelligence. Volume 38:Number 2(2022)
• Page Start: 586
• Page End: 609
• Publication Date: 2021-11-17
• Subjects: adaptive partitioning -- deep neuro‐fuzzy network -- forgery detection -- local Gabor XOR pattern -- rectangular search; Artificial intelligence -- Periodicals; Computational linguistics -- Periodicals; 006.3
• DOI: 10.1111/coin.12484
• Languages: English
• ISSNs: 0824-7935

Abstract

Abstract: The emergence of photo editing applications, like Adobe Photoshop, has manipulated the operation of digital images into a simple task. However, these manipulations of images misrepresent the content of the original image for misleading the public. Various copy move forgery detection techniques are developed, but these show less robustness on the image with noise and blurring. This article develops an optimization‐driven deep learning technique for image forgery detection. The purpose is to develop a copy‐move image forgery detection technique using a deep neuro‐fuzzy network and a newly developed optimization algorithm. Here, adaptive partitioning is adapted using a rectangular search for splitting the image into different parts. In addition, the features like local Gabor XOR pattern and Texton features are extracted from the partition. Furthermore, the forgery is detected using the deep neuro‐fuzzy network. Finally, the deep neuro‐fuzzy network training is performed using the proposed multi‐verse invasive weed optimization (MVIWO) technique. The proposed MVIWO method will be newly designed by integrating the multi‐verse optimizer and invasive weed optimization technique. Thus, the copy‐move image forgery detection is effectively performed using the proposed MVIWO‐based deep neuro‐fuzzy network. The developed MVIWO‐based deep neuro‐fuzzy network offers superior performance with the highest specificity of 93.54%, highest accuracy of 94.01%, and highest sensitivity of 97.75%.