
Copy-Move Forgery Detection Based on Pyramid Correlation Network
Author(s) -
Liang Peng
Publication year - 2021
Publication title -
converter
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.104
H-Index - 1
ISSN - 0010-8189
DOI - 10.17762/converter.109
Subject(s) - pyramid (geometry) , benchmark (surveying) , computer science , block (permutation group theory) , artificial intelligence , correlation , pattern recognition (psychology) , image (mathematics) , representation (politics) , spatial correlation , computer vision , mathematics , geography , cartography , telecommunications , geometry , politics , political science , law
Block-based image copy-move detection algorithms disregard the spatial layout of the features, leading to the poor detection performance under small-region tampering samples. Therefore, we propose a pyramid correlation network (PCNet) for copy-move forgery detection, whose goal is to obtain rich and detailed image representation via a pyramid cascaded correlation architecture. Experimental results show that PCNet outperforms the comparison algorithm on USCISI, CASIA and CoMoFoD data sets. Compared to the benchmark model BusterNet, F1 scores of PCNet has increased by 33.84% and 30.62% on CASIA CMFD dataset and CoMoFoD dataset respectively.