Unsupervised colour image segmentation using dual-tree complex wavelet transform

In this paper we present an effective unsupervised colour image segmentation algorithm which uses multiscale edge information and spatial colour content. The multiscale edge information is extracted using the dual-tree complex wavelet transform. Binary morphological operators are applied to the edge information to detect seed regions which are large enough to exclude boundaryonly regions. The segmentation of homogeneous regions is obtained using region growing followed by region merging in the CIE Lab colour space. We also present an edge-preserving smoothing filter as a pre-process for the algorithm. We compare our algorithm with state-of-the-art algorithms and show its superior performance.