Texture Image Segmentation Based On Stationary Wavelet-based Nonsubsampled Contourlet

Recently, the application of Multi-scale geometric analysis in the image segmentation has been more and more concerned. Contourlet is of good performance in dealing with the singularity of the image characteristics, but there are shortcomings of lacking of redundancy and having no translational invariance. Stationary Wavelet-Based Nonsubsampled Contourlet transform (SWTNSCT) is proposed. It is a transform based on stationary wavelet transform and non-sampling direction filter. It is an image analysis tool incorporating both multi-scale of Wavelet and multi-direction of Contourlet, and it is fully translation-invariant and having strong relevance. It is suited to image segmentation than Contourlet. In this paper, a comprehensive study of Stationary Wavelet-Based Nonsub-sampled Contourlet transform in image segmentation applications is made, and the principal tasks are as follows:(1) Stationary Wavelet-Based Nonsubsampled Contourlet is applied to the texture image segmentation. It is full use of the multi-scale, multi-directional resolution of Stationary Wavelet-Based Nonsubsampled Contourlet wavelet transform, and the texture image segmentation is realized by the method of FCM algorithm. The algorithm is full use of more direction information and statistical characteristics of Stationary Wavelet-Based Nonsubsampled Contourlet, and it gets good segmentation results. The simulation experimental results show that the method is better in the uniform region than Wavelet and Contourlet transform.(2) The image segmentation algorithm of Stationary Wavelet-Based Nonsub-sampled Contourlet transform with Immune Clone Selection Algorithm is proposed, mainly for much more texture features extracted by Stationary Wavelet-Based Nonsub-sampled Contourlet transform, and some of the direction sub-bands contain less amount of information. Therefore, the paper uses the Immune Clone Selection Algorithm making use of its global optimization, to eliminate the texture features against segmentation result, and select appropriate texture features to realize the image segmentation. After joining the Immune Clone Selection Algorithm, the computational complexity is reduced, and the accuracy of image segmentation is improved. Trough the simulation experiment, we can see the method result mixing the Immune Clone Selection Algorithm is better than before.