Abstract | ||
---|---|---|
The primary goal of this dissertation is the modeling and detection of compound objects, such as harbors and golf courses, in remotely sensed geospatial images. Toward this goal, this dissertation makes two important contributions: (1) it demonstrates the potential of frequency-domain texture analysis for model-driven detection of geospatial objects, and (2) it addresses the problem of learning appearance models for objects from their examples. The complexity of geospatial image content present obstacles in using purely spatial (pixel) domain methods for describing objects. In this dissertation, the structure of objects is efficiently described using Gabor filter-based texture analysis, which incorporates information from both the spatial and frequency domains. The use of texture motifs, or characteristic spatially recurrent patterns, is proposed for modeling and detecting geospatial objects. Three approaches are described in this dissertation for learning texture motif representations from object examples and detecting objects based on the learned models. The first approach is a two-layered representation that first learns the constituent "texture elements" of the motif and then the spatial distribution of the elements. In the second approach, the texture elements of a motif are learned as the states of a hidden Markov model (HMM), and the state transitions of the model describe the spatial arrangement of the elements. The third approach addresses the problem of detecting objects with quasi-periodic texture motifs, by analyzing the relations among the characteristic scales and orientations of these patterns. Experimental results demonstrate the effectiveness of the above approaches in detecting compound geospatial objects. |
Year | DOI | Venue |
---|---|---|
2006 | 10.1109/TGRS.2006.881741 | Geoscience and Remote Sensing, IEEE Transactions |
Keywords | Field | DocType |
image texture,object detection,remote sensing,geospatial objects,image tiles,large aerial image datasets,model training,object detection,object model,spatially recurrent pattern,texture motif,texture-motif model,Geospatial object,object detection,object model | Geospatial analysis,Computer vision,Object detection,Image texture,Remote sensing,Object model,Aerial image,Artificial intelligence,Mathematics | Journal |
Volume | Issue | ISSN |
44 | 12 | 0196-2892 |
ISBN | Citations | PageRank |
0-542-47645-2 | 33 | 2.48 |
References | Authors | |
38 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Sitaram Bhagavathy | 1 | 114 | 9.82 |
B. S. Manjunath | 2 | 7561 | 783.37 |