A comprehensive evaluation procedure for copy-move forgery detection methods: Results from a systematic review
In the current age, the wide use of digital images has led to the manipulations of content that misrepresent information with malicious goals. This issue demands the requirement of digital image investigation to authenticate the source and certify the trustworthiness of images. One image manipulatio...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Published: |
Springer
2022
|
| Subjects: | |
| Online Access: | http://eprints.um.edu.my/43008/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In the current age, the wide use of digital images has led to the manipulations of content that misrepresent information with malicious goals. This issue demands the requirement of digital image investigation to authenticate the source and certify the trustworthiness of images. One image manipulation technique is called copy-move forgery (CMF), which duplicates one or more regions in an image before it is pasted to another location within the same image. In this paper, a systematic review is conducted to assess the performance evaluation techniques implemented by current CMF detection methods' approaches. Five research questions are generated to find and solve the related issues on the evaluation levels. At present, CMF detection performance is evaluated either through image-level, pixel-level, or both level evaluations. Image-level evaluation identifies an image either as forged or not while all images in pixel-level evaluation are treated as forged images to localise the CMF regions. The study shows that both image and pixel-level evaluations are dependent and must be incorporated together to ensure a fair evaluation is conducted. A comprehensive evaluation procedure that covers both evaluations is proposed as a guide to future research. The procedure is then examined with seven state-of-the-art CMF detection methods based on keypoint, block, and combination approaches using three available CMF datasets that consist of multiple CMF attacks. The results are measured using multiple F-score values: image scores, pixel scores, and a multiplication of both of these to get the overall percentages of the detection. The results show that the block-based approach is able to obtain the highest percentages of detection in almost all cases of attacks in CMF. |
|---|