Visual steganalysis of LSB-encoded natural images

Contemporary steganographic systems encode hidden messages inside the least significant bit layers of colour natural images. The presence of these messages is difficult to detect through statistical attacks. This study examined whether humans could detect steganography in natural images using a controlled 2AFC discrimination task. While d'>l was observed for colour layers 3-8, layer 1 had a negative d'. Thus, layer 1 embedding is highly resistant to visual attack, since observers were more likely to indicate the presence of steganography in the control image than the embedded image.