Conference/Proceedings2nd Workshop on Immersive Communication and Broadcast Systems (ICOB '05)
Start date27.10.2005
End date28.10.2005
AddressBerlin, Germany
Author(s)Jang-Heon Kim, Matthias Kunter, Thomas Sikora
TitleDepth Diffusion Objects (DeDiO) - A Seamless Object-based Approach for TV Applications
AbstractThis paper proposes a novel and seamless object-based method
for visual contents fusion called “Depth Diffusion Object
(DEDIO)”. It employs depth-based object segmentation in multiview scenes and allows the composition of new scenes
conveying a natural image impression. Using spatial anisotropic diffusion, the homogeneous regions of a scene which have similar depth are smoothly regularized following the spatial variation while discontinuities are preserved. The object’s shape is automatically extracted by the depth range. In order to fuse the object with new background images we use membrane harmonizing applying directional constrained diffusion to remove visual seams and blend object, i.e. the remaining background area in the segmented object near the shape boundary. Our system automatically maintains the seamless object composition quality, even if the object is merged with new scenes, having different lightening conditions.