A Parsimonious Self-Organizing Neuro Fuzzy System for Image Classification with Feature Ranking and Wavelet Transformations

A parsimonious self-organizing neuro fuzzy system for image classification is proposed in this paper. Self-organizing neuro fuzzy system is designed by synthesizing fuzzy clustering and support vector machine which is applied to classification problems. Fuzzy clustering decides the number of rules instead of support vectors resulting in smaller number of rules. For enhancement the most appropriate features relevant to the classification task are identified by feature ranking. Feature ranking and training are performed simultaneously in an integrated way resulting in good generalization performance. Instead of taking the whole image, the Haar discrete wavelet transform (Haar DWT) of the image is considered and only the first sub band of the image after applying Haar DWT is taken for feature extraction. We use Haralick and Texture Spectrum features as image content descriptors and efficiently employ them to retrieve images.