Multi-dimensional histogram method using multi-spectral images

In this paper, we present a multi-dimensional histogram method for classifying multi-spectral image pixels into a particular category. The proposed method consists of the following four steps: create a multi-dimensional histogram its the database of a particular category by collecting category information which is given by researchers, classify unknown image pixels by comparing them with the database, modify the result of the classified pixels, and merge the modified information into the database and continue to the second step. Thus, the database hits been updated and upgraded considerably. Therefore, we can automatically classify unknown image pixels into the category as the researchers can manually do. Since researchers use not only spectral information but also geometrical information for classifying image pixels, a pixel sometimes exists in some categories at the same time. It makes the accuracy of the result lower and exists among boundaries of the categories. The proposed method can almost prevent the pixel from decreasing the accuracy by making a pixel exist in only one category. We applied the proposed method to classify 92 images of NOAA AVHRR into the sea category. It succeeded to achieve the classification accuracy of 94% on average.