A study on detecting image hiding by feature analysis
Guo-Shiang Lin, Wen‐Nung Lie · 2002
A feature analysis scheme based on gradient energy in the spatial domain and Laplacian parameter in the DCT domain is proposed for identifying the existence of secret or proprietary data hidden in an image. The applications may be for an interceptor which is responsible for examining and sifting out those suspicious images used to convey secret data or messages. The spatial feature describes the phenomenon that continuity or smoothness of gray levels between adjacent pixels is destroyed by the disturbing noise and this will increase the gradient energy. The second feature characterizing the Laplacian distribution of DCT coefficients is used to distinguish the original from the modified images. By experiments, it is found that the proposed features can achieve a high detection rate (over 90%) at a low false alarm rate (regardless of the data embedding schemes, spatial or DCT domain, adopted) provided that the computed features are suitably scaled or normalized.