Non-Parametric Classification of Remotely Sensed Multispectral Image Data by Means of Matrix Representation of Multidimensional Histograms
Department of Electronic Engineering, Faculty of Engineering, Gunma University, Kiryu-shi, Gunma, 376 Japan
Received:November 26, 1993Accepted:December 10, 1993Published:February 20, 1994
Keywords:Land use map, Maximum likelihood method, Category classification accuracy, Spectral-spatial image
The computer framing of land use maps using remotely sensed multispectral image data is identical with pattern classification for spectral reflectance of objects on earth's surface. In particular, the classification by the maximum likelihood method is the most popular method because it theoretically gives the highest correct classification rate on the condition that the statistical distribution of the image data be normal. However, the histogram of real image data is not a normal distribution. Actual histograms show the proper distributions to classes. This fact means that a histogram gives a spatial property of the class statistically. This paper described a newly developed non-parametric method by means of the matrix representations of multidimensional histograms and subimages.
Cite this article as:M. Inamura, “Non-Parametric Classification of Remotely Sensed Multispectral Image Data by Means of Matrix Representation of Multidimensional Histograms,” J. Robot. Mechatron., Vol.6 No.1, pp. 42-50, 1994.Data files: