Image Classification using Chip Mining Approach for Remote Sensing Imagery

摘要

In this paper we present a concept of indiscernibility relation and upper approximation from rough set theory in image chip mining technique for image classification. Chip library of 2×2 pixels is created for our defined terrain classes. Window of 2×2 pixels is rotated in the input image and indiscernibility relation is computed with the chips in the library. Depending on indiscernible relation the window is leveled according to its class. Hence, it can be said that window certainly belongs to upper approximation. The final segmental image is formed after the region fusion process using the statistical criteria. The method integrates machine learning paradigm, especially learning from example techniques. Issue for construction of sufficient image chips and representation of it, to make an efficient and intelligent machine is also discussed. This work mainly address for classification of remote sensing data comprising the concept of rough set theory contrary to traditional work on image classification. Result is presented in the end of paper.