Content Based Retrieval for Remotely Sensed Imagery

We present a framework for content based retrieval (CBR) of remotely sensed imagery. The main focus of our research is the segmentation step in CBR. A bank of gabor filters is used to extract regions of homogeneous texture. These filter responses are utilized in a multiscale clustering technique to yield the final segmentation. Novel area morphological filters are utilized for the purpose of scaling. The resultant segmentation yields regions that are homogeneous in terms of texture and are significant in terms of scale. These regions are used for the purpose of extracting shape and textural features (on a global and local basis) that provide important similarity cues in CBR of remotely sensed imagery. In comparison to solutions which use region merging, the segmentation from the texture / scale space does not require heuristic postprocessing, nor knowledge of the number of significant regions.