An Image Classification Based on New Similarity Measure in Rough Sets
Takehiko Matsuura, Syoji Kobashi, Katsuya Kondo, Yutaka Hata
Proc. of 1st Int. Conf. on Fuzzy Systems and Knowledge Discovery, CD-ROM, (2002)
This paper proposes an image classification method using 'value reduct,' which is a concept of rough sets. In this method, attribute values of each pixels of an image are given by using K-means clustering, and the attribute values divide the image into many regions. By applying the value reduct to the attribute values, dissimilarities between regions are calculated. Dissimilarities between regions make initial equivalence relations. Similarities between regions are calculated based on the relations. Finally, this method forms final equivalence relation between regions based on similarities, and classifies them into some regions according to the relations. The performance of the algorithm was evaluated by applying it to an artificial generated image and a human brain MR (magnetic resonance) slice image.