Recaptured photo detection using specularity distribution

Detection of planar surfaces in a generic scene is difficult when the illumination is complex and less intense, and the surfaces have non-uniform colors (e.g., a movie poster). As a result, the specularity, if appears, is superimposed with the surface color pattern, and hence the observation of uniform specularity is no longer sufficient for identifying planar surfaces in a generic scene as it does under a distant point light source. In this paper, we address the problem of planar surface recognition in a single generic-scene image. In particular, we study the problem of recaptured photo recognition as an application in image forensics. We discover that the specularity of a recaptured photo is modulated by themesostructure of the photo surface, and its spatial distribution can be used for differentiating recaptured photos from the original photos. We validate our findings in real images of generic scenes. Experimental results show that there is a distinguishable feature of natural scene and recaptured images. Given the definition of specular ratio as the percentage of specularity in the overall measured intensity, the distribution of specular ratio image’s gradient of natural images is Laplacian-like while that of recaptured images is Rayleigh-like.