Object segmentation for Z-keying using stereo images

We propose a robust object segmentation framework exploiting multiple cues such as shape, intensity (color) and depth. Though over last few decades, various segmentation, schemes have been developed, those schemes based on intensity and motion information have well-known disadvantages. To alleviate those problems we take into account depth information using MRF/GRF framework. The experimental results show the effectiveness of the proposed framework by clearly separating "objects of interest" from real, rather than blue-screen, scenes. The proposed scheme will be a key part for wide scope of applications requiring object-based functionalities as well as Z-keying for photo-realistic mixed reality.