Object-based Depth Post-processing
This project provides a novel spectral analysis method for 3D images on the basis of the concept of objects. Based on the analysis, an automatic 3D image correction is also proposed. The analysis involves the construction of 3D object boundaries from Hilbert Huang Decomposition. A 3D grouping adaptation algorithm tries to match the concept of 3D objects for the correction. Even for partial occluded objects, the grouping algorithm is still working. The experimental results in Fig. 4 show that the correction improves quality of 3D image signals.
Fig. 4. Correction for Depth (a)(b) Original Stereo Pair (p.s. (a):left (b):right) (c) Matched Depth for Left Image (d) Corrected Depth for Right Image (e) Anaglyph of Depth from Stereo Matching (f) Anaglyph of Corrected Depth