Start Date
10-12-2017 12:00 AM
Description
Modern societies are clearly en route towards digitalization. Natural scene images particularly could provide many value added applications and services. In this paper, we address the challenges that arise with pavement detection in rural, urban, and unstructured scenes. We use an approach combining region similarity, split and merge, and pairwise assignment, to merge image regions according to a homogeneity criteria, and group pavement regions in images. Homogeneity criteria is based on color and texture feature types. In successive steps a distance matrix based on varying kernel is created. The approach is numerically simple, yet retains the ability to merge similar regions. Assignment with Hungarian method is used to achieve hierarchical region merging. As a result, pavement area grouping could be handled. Evaluations on relevant datasets show that our approach allows successful merging and recognition of image regions belonging to the pavement area for various scenes and surface types.
Recommended Citation
Chatterjee, Sromona; Hildebrandt, Björn; and Kolbe, Lutz Maria, "Understanding the Scene Data- Pavement Area Grouping in Images" (2017). ICIS 2017 Proceedings. 19.
https://aisel.aisnet.org/icis2017/DataScience/Presentations/19
Understanding the Scene Data- Pavement Area Grouping in Images
Modern societies are clearly en route towards digitalization. Natural scene images particularly could provide many value added applications and services. In this paper, we address the challenges that arise with pavement detection in rural, urban, and unstructured scenes. We use an approach combining region similarity, split and merge, and pairwise assignment, to merge image regions according to a homogeneity criteria, and group pavement regions in images. Homogeneity criteria is based on color and texture feature types. In successive steps a distance matrix based on varying kernel is created. The approach is numerically simple, yet retains the ability to merge similar regions. Assignment with Hungarian method is used to achieve hierarchical region merging. As a result, pavement area grouping could be handled. Evaluations on relevant datasets show that our approach allows successful merging and recognition of image regions belonging to the pavement area for various scenes and surface types.