Feature Discrimination in Large Scale Satellite Image Browsing and Retrieval

Feature discrimination is a critical component in large scale satellite image browsing and retrieval, as well as in other areas of remote sensing. As many satellite images are multi-spectral, the spectral information should be exploited for efficient browsing and retrieval. In this aspect, color is a suitable candidate to represent the spectral contents. This paper presents an efficient algorithm on feature discrimination for browsing and retrieving of large-scale satellite imagery using color features. Our main focus is on color modeling of interest region based on appropriate color space. The description of color features using spherical influence fields allows unrealistic or biased color prototypes to be removed from the color model. And the proposed color model is capable of discriminating the interest region from the image background. Experimental results demonstrated that interest region was successfully discriminated from the complex scene using the proposed color model in the L*a*b color space.