DEPTH MAPS CODING WITH ELASTIC CONTOURS AND 3D SURFACE PREDICTION

Marco Calemme**, Pietro Zanuttigh*, Simone Milani*, Marco Cagnazzo**, Beatrice Pesquet-Popescu**

* Multimedia Technology and Telecommunications Laboratory, University of Padova

** Groupe Multim├ędia, Telecom Paristech, Paris, France

 

 

This page contains two examples of view interpolation performed with depth data compressed with the proposed approach and with HEVC. The 3D-HTM software renderer version 6.2 has been used.

The first table shows the reconstruction of 5 frames of view 7 of the "lovebird" sequence reconstructed from views 6 and 8.
Depth data compressed at around 0.008 bpp with both the proposed approach and with HEVC have been used.
Notice how there are relevant artifacts in proximity of the people edges in the ones produced with HEVC while the ones coded with our approach have much smaller edge artifacts.

Our Approach
HEVC
Frame 



In the second example 5 frames of the views 1 and 5 of the "undo dancer" sequence have been used to reconstruct view 3.
Depth data compressed at around 0.005 bpp with both the proposed approach and HEVC have been used.
Again HEVC coding leads to more edge artifacts in proximity of the person edge (in this sequence they are harder to see, click on the images to enlarge them and zoom the image in proximity of the person edge).

Our Approach undo_dancer  undo_dancer  undo_dancer  undo_dancer  undo_dancer 
HEVC undo_d_hevc  undo_d_hevc  undo_d_hevc  undo_d_hevc  undo_d_hevc 
Frame  1 2 3 4 5

If you are interested in our research you can visit our websites: http://lttm.dei.unipd.it , http://www.tsi.telecom-paristech.fr/mm