A new images segmentation scheme, which is based on combining technique of fractal dimension and self-organization neural network clustering, was presented in this paper. As we know features extracting is a very important step in image segmentation. So, in order to extract more effective fractal features from images, especially in the remote sensing images, a new image feature extracting and segmentation method was developed. The method extracts fractal features from a series of images that are obtained by convolving the original image with various masks to enhance its edge, line, ripple, and spot features. After that a 5-dimension feature vector are procured, in this vector each element is the fractal dimension of original image and four convolved images. And at last, we segment the image based on the strategy that combining the nearest neighbor classifier with self-organization neural network. Applying the presented algorithm to several practical remote sensing images, the experimental results show that the proposed method can improve the feature description ability and segment the images accurately.

Date Published: 25 April 2008
PDF: 8 pages
Proc. SPIE 7001, Photonics in Multimedia II, 70010L (25 April 2008); doi: 10.1117/12.780160

Published in SPIE Proceedings Vol. 7001:
Photonics in Multimedia II
Ari Tervonen; Frank Möllmer, Editor(s)