Abstract
The texture gives useful information for dividing a picture into segments, and the analysis by using this information is called a texture analysis. Here two fundamental problems will be encountered; namely,
1) how this texture is characterized or modeled, and 2) how the demands on the class separability and on the segmentation accuracy are compromised.
For the first problem this paper gives a mathematical model of texture to descrive the 1st order density function of pixcell grayness and the autocorrelation function. By synthesizing a texture sample following this model, the model fitness is investigated visually.
For the second problem is given a classification method which uses minimum pixcells for classification through the above mentioned mathematical model under the condition that the class separability must be greater than a given value.
Simulation of this classification method resulted to provide sufficient class separability without much loss of the segmentation accuracy.
