1999 年 38 巻 3 号 p. 44-59
Authors propose a parameter estimation method for a mixel problem in satellite images with coarse resolution. The method is based on an assumption that mixel class has its own probability distribution, and it can be calculated through convolution between pure class probability distributions near boundaries. In this paper, firstly the mechanism of mixel occurrence is explained through simulated images. Then, the proposed method for mixel decomposition is described. The result of the experiment using simulated images demonstrates that our method can decompose image's histogram and estimate each parameter of class probability distribution, especially mixel class ones, more accurately than a conventional method in terms of Kullback-Leibler entropy.