With the advances in image manipulation software, it has become easier to manipulate digital images. These manipulations can be used to increase image quality, but can also be used to depict a scene that never occurred. One of the purposes of digital image forensics is to identify such manipulations. There is however a lack of research on the detection of manipulated stereoscopic images. Stereoscopic images are images which create an illusion of depth for the viewer by showing an image pair that correlates to a person’s left and right eye.
This dissertation investigates how depth information can be used to detect stereoscopic image manipulations. Two techniques were developed and tested through experimentation.
The first technique used disparity maps to highlight large areas without internal depth. These areas can be the product of non-stereoscopic to stereoscopic splicing techniques. Experimentation results showed that areas without internal depth can be detected. However, the detected areas can be the product of natural occurrences in images and not only non-stereoscopic to stereoscopic splicing. Post investigation of detected areas is thus required to verify the results. The second technique used a derived formula to determine the distance an area will lose internal depth. Experimentation results showed that the formula is fairly accurate. This information can be used to aid the detection of internal depth inconsistencies in stereoscopic images. These inconsistencies can arise due to stereoscopic image splicing or other image manipulation techniques that may modify the internal depth of a stereoscopic image.