Interactive analysis of large scale multi-spectral images using a Hilbert curve

There have been some new developments of an interactive analysis for the multi-spectral images. Recently the authors have proposed an interactive analysis method for classification using a Hilbert curve which is a one-to-one mapping and takes a neighborhood between N-dimensional space and one-dimensional space into consideration. In order to analyze large scale multi-spectral images, we divide a large scale image into subimages which can be analyzed using our proposed method. A problem is that after classifying one of the subimages, how we classify the rest of the subimages using this result effectively. We present a solution of this problem using a tree structure expression. We assign a reliability measure to each pixels on the rest. The reliability measure is based on a distance from a center of a cluster, and the center is considered occurrence information. For the low reliable data, we apply our interactive analysis method for classification again. In the experiment using a LANDSAT image data, We confirmed the effectiveness of the reliability measure because category boundaries on the rest have lower reliability.