A technique for fake 3D (2D-to-3D converted) video recognition

In this paper, we propose a technique for the automatic recognition of "fake" stereoscopic videos/movies i.e., videos which result from classic 2D videos through a 2D to 3D conversion process. Essentially, the proposed technique distinguishes between 2D movies converted to 3D and real stereoscopic ones. It is based on the difference in sharpness around foreground objects in a converted stereo frame pair caused from the in-painting step that takes place after the generation of the right frame (rendered view) from the left frame (source view). The two variants of the algorithm, one utilizing a two-class Support Vector Machine and another one that follows a threshold based classification approach, use a sharpness metric evaluated on a stripe created around foreground objects such as human figures. Experimental evaluation of the proposed algorithm, which can serve as 3D quality characterization tool, is conducted on several stereoscopic movies with very promising results.