A New Method Based on Fused Features and Fusion of Multiple Classifiers Applied to Texture Segmentation

Texture image segmentation consists of two stages: feature extraction and classification. The new method advanced in this paper fuses the log-gabor filter and DCT features in the first stage, then uses the fusion of fuzzy c-means (FCM) and support vector machines (SVM) classifier to cluster the fused feature sets. The fused feature sets produce higher feature space separations, and the fusion of multi-classifiers performs the better clustering effect. The new method is demonstrated to produce higher segmentation accuracies relative to the individual feature and individual classifier, as well as outperform individual feature for noisy images with different noise magnitudes. The fused features and classifier fusion are advocated as means for improving texture segmentation performance.