A Simple Semi-automatic Technique for 2D to 3D Video Conversion

3D movies and 3D TV have become more and more popular, but they may be hampered by the lack of 3D contents. One possible solution of this problem is to convert the existing monocular videos into 3D versions. This paper presents a semi-automatic technique for 2D to 3D video conversion. Our method segments key frames first through interactive watershed algorithm, and then the partitions are manually assigned with their corresponding depth values. For non-key frames, we use interpolation approach to obtain initial contours of the different partitions, and then adjust them with snake model. Based on the computed depth image, the left view and the right view are generated through rendering followed by hollows' filling, and the 3D image of each frame is synthesized, and then the 3D video is integrated. Our approach can simplify the operation of key frames, and improve the tracking accuracy of non-key frames. Experimental results on a set of monocular movies have proved that our algorithm is a practical technique for 2D to 3D conversion.