Wavelet features for information mining in remote sensing archives

Feature selection and extraction is an initial stage of image information mining, hence it is the vital step in the whole process. Content and semantic based interactive mining systems describe remote sensing images by means of relevant features. Using global features based on color, shape, and texture, these systems possess a rich semantic value, but they fail to capture the local properties of the image. To overcome this limitation, a number of region-based retrieval systems have been proposed. Existing region-based retrieval systems for information mining in remote sensing image archives extract primitive features based on color, texture, and shape from the segmented homogenous region and retrieve images based on similarity regions. The texture features used are based on the Spatial Gray Level Dependency (SGLD) matrices. The region-based image retrieval techniques using wavelet transforms have recently gained momentum in multimedia image archives. Inspired by the successful application of wavelet transform techniques in other imagery archives, this paper proposes to perform image segmentation using color and texture features from the wavelet coefficients for the region-based retrieval in remote sensing image archives. The feasibility study shows that the wavelet technique could be used in retrieval of images from the remote sensing archives.