Consistent line clusters for building recognition in CBIR

This paper introduces a new mid-level feature, the consistent line cluster for use in content-based image retrieval. The color, orientation, and spatial features of line segments are exploited to group them into line clusters. The interrelationships among different clusters and the intra relationships within single clusters are used to recognize and roughly locate buildings in photographic images. Experiments are performed on a database of color images of outdoor scenes.