 |
|
|
| project results: |
Fully Bottom up Interpretation of Man-Made ScenesMohammad Jahangiri, Daniel Heesch, Robby Tan and Maria PetrouCommunication and Signal Processing Group Imperial College London October, 2009 A fully bottom up interpretation module for interpreting images of man-made scenes, in particular buildings is proposed. For this a three layer interpretation system was designed which in the lowest level meaningful regions which are salient in gradient are detected, using the so called blob detector algorithm. After identifying the blobs, in the second layer, higher order structures such as repeated patterns, building facade and also parts of the environment like sky and ground are segmented. A unified map was then estimated using different segmented regions so that each segment contains a meaningful part of the scene. In the third layer, the segmented regions are labelled using a MRF based contextual classifier. This classifier was employed as the buildings and their subparts are inconsistent in the low-level features. The accuracy of the proposed scheme was evaluated using the 12-class dataset of eTRIMS.
References
|
