Image Splicing Detection Using Depth-Wise Convolution Neural Network

Images play a pivotal role in documenting real-life events. With the rapid evolution of digital technology, there has been a significant increase in both the creation and dissemination of photographs. The accessibility of picture editing software has simplified the process of altering images, thereby reducing the time, costs, and expertise needed to create and manage visually manipulated content. Unfortunately, digitally altered photographs have become a primary medium for disseminating misinformation, which affects individuals and society at large. Consequently, the need for effective methods to detect and identify forgeries is more pressing than ever. One prevalent form of picture fraud, image splicing, has been thoroughly examined. In this study, we present a Depth-Wise Convolutional Neural Network (DWCNN) model specifically designed to accurately detect spliced forged images. By converting input RGB images to the HSV color space, known for its ability to withstand color and lighting variations, our model achieves high accuracy in identifying manipulated images. Furthermore, our proposed model is lightweight, based on the MobileNet architecture with seven bottleneck blocks, making it suitable for a wide range of scenarios with constrained resources. To evaluate the model's performance, we tested it on the CASIA v1.0 and CASIA v2.0 datasets. Our model accurately identified forgeries with 99.23% accuracy on the CASIA v1.0 dataset and achieved a remarkable accuracy of 99.37% on the CASIA v2.0 dataset.