Blind forgery detection using enhanced mask-region convolutional neural network

This work proposes a DL model for locating and classifying the forgery images. The proposed work has stages like pre-processing, feature extraction, segmentation, localization, and forgery detection. The input RGB images are converted into an YCbCr colour model in the pre-processing stage. Due to the size of the blocks, the processing time may be increased. Hence, the input images are split into overlapping blocks to reduce the time complexity. Here, a series of residual blocks are utilized for feature extraction, segmentation, and localization. Hybrid DL model Enhanced Mask-RCNN carries out this process. The Enhanced Mask-RCNN integrates the residual network with Mask-RCNN (Mask-region convolutional neural network). In this hybrid network, the residual network is used to extract the features, and the Mask-RCNN is used to segment, locate, and detect the tampered region. Further, the neural network weights are optimized by sandpiper optimization (SO) to enhance the recognition accuracy. The performance of a proposed forgery detection model is compared with three benchmark datasets and attained better accuracy of 0.991, 0.997, and 0.997 on the GRIP, Coverage, and CASIA-V1 datasets, respectively. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.