Segmentation of color textured images using dual tree complex wavelet features and fuzzy clustering

In this paper, we have proposed a novel approach for segmentation of color textured images using dual tree complex wavelet transform (DTCWT). The choice of DTCWT is proven for its attractive properties such as shift invariance, good directional selectivity, limited redundancy and efficient computation. The image is first decomposed into 16 sub bands by applying one level DTCWT in a separable manner without down sampling. Texture features are extracted from these sub bands by estimating in each sub band, the local energy around each pixel over a small neighborhood. To offer dimensionality reduction in feature space, a decisive criterion called mutual information is used to remove irrelevant or redundant features. Having obtained the feature images, we emphasize on the feature representation by labeling each pixel independently using fuzzy clustering. The segmentation results have demonstrated the effectiveness of the proposed method.