RECOLORED IMAGE DETECTION USING A DEEP DISCRIMINATIVE MODEL

Abstract

With the rapid advancement of image editing tools and deep learning technologies, digital image manipulation has become increasingly sophisticated, making it difficult to distinguish between original and altered images. Recolored image manipulation, where colors of an image are intentionally modified without changing structural content, poses a significant challenge in image forensics. This project proposes a deep discriminative model for detecting recolored images by analyzing subtle inconsistencies in color distribution and pixel relationships. The proposed system utilizes deep learning techniques, particularly Convolutional Neural Networks (CNNs), to extract high-level features from images. The model is trained to differentiate between original and recolored images by learning discriminative patterns that are not easily visible to the human eye. The approach focuses on color channel inconsistencies, histogram variations, and pixel correlation features to improve detection accuracy. A labeled dataset containing original and recolored images is used for training and evaluation. Experimental results demonstrate that the deep discriminative model achieves high accuracy in identifying recolored images compared to traditional image processing methods. The system is robust against various recoloring techniques and provides reliable performance in realworld scenarios. However, challenges such as dataset diversity and computational requirements remain. This research contributes to the field of digital image forensics by providing an effective solution for detecting subtle image manipulations.