Linear fuzzy space polygon based image segmentation and feature extraction

In this paper we propose a new model and algorithm for region segmentation and feature extraction from 2D images containing imprecise regions. Region modeling is done in two phases. In the first phase a region is represented as a classical fuzzy set, and in second phase the obtained fuzzy set is approximated by a fuzzy polygon, another fuzzy set whose borders are represented as an array of fuzzy points in linear fuzzy space. Membership functions for the fuzzy set in the first phase are represented by feed forward neural network trained on the set consisting of pixels´ feature vectors. The feature vector model is formed based on 2D wavelet transformation of pixel´s neighborhood. Utilization of the model and algorithm is demonstrated through the example of calculating region diameter in DICOM 2D medical images.