TIDE: tampered image detection using mutual information

Abstract

In this paper, we study the problem of detecting tampering in images without having any prior knowledge of image and its content. The features are designed to classify the given image as raw image (cover image) or image containing hidden data (stego image) embedded into original image. Two set of features are designed – one based on histogram of image and other based on information theoretic measure such as mutual information. Histogram of image is analysed using short-time Fourier transform (STFT) and features based on centre of mass (COM) in frequency domain is designed. Statistical dependency between adjacent pixels in natural images is quantified using Mutual Information. The observations made in our analysis provide some interesting observations on image tampering detection using features based on STFT and mutual information and short-time Fourier transform. We have performed the experimental result using the coral database containing 10,000 images and observed 85.71% classification accuracy which is a significant improvement over the previously reported techniques.