An Improved Segmentation Algorithm for Textured Color Images Using Dual Tree Complex Wavelet Derived Features In Histogram Thresholding Techniques

This paper proposes an improved texture segmentation algorithm based on the features derived from Dual Tree Complex Wavelet Transform (DTCWT) which is proved to be efficient for texture description. The dual tree introduces limited redundancy, approximate shift invariance and directional selectivity while preserving perfect reconstruction and computational efficiency.DTCWT is applied on the three components of the input color image.Co occurrence features are computed for the resultant sub images.Then, the sub image which has the maximum energy is selected for which local homogeneity is calculated. Various histogram thresholding techniques are applied separately on the resultant homogeneity histogram. The experiments of segmentation provide more encouraging results for textured color images using peak finding algorithm than those based on Mean shift and Otsu multi thresholding algorithms. The results obtained using a set of real world colored textures demonstrated the usefulness of wavelet features in color texture image segmentation.