Subimage Selection from Multiple Image on Eigenspace for Clustering

Abstract

Clustering of multiple image such as multispectral and hyperspectral image has some challenges. First, the computational complexity increases in proportion to the number of multiple image. Second, the performance is decreased by mixed improper images. Therefore, dimensionality compression and dimensionality reduction of multiple image are needed for the computational complexity decrease and the accuracy improvement of clustering. Then, this technical report marks on dimensionality reduction of the multiple image, and proposes a method to select subimage from multiple image on the eigenspace. To validate the effectiveness of the proposed method, selection of subimage is performed using Landsat TM multispectral image. A landcover classification map which is created as the result of clustering is compared with aerial photo.