Statistical shape based multispectral image retrieval extracting Power Spectrum Vectors

This paper reports a novel approach to statistical shape-based multispectral image retrieval. The proposed method starts with the definition and calculation of a statistical shape feature called power spectrum vector (PSV). Then independent component analysis (ICA) is used to extract independent PSV components. Later on, the PSV value for each image is presented by a linear combination of these independent components. In this way, a weight vector is calculated for each image in the system. These weight vectors are effectively used in the statistical shape-based retrieval process. The proposed method is implemented and tested on a set of LANDSAT multispectral images from variant sceneries. Finally, experimental results and evaluation for this approach are presented.