Segmentation of Noisy Images Using Improved Distance Regularized Level Set Evolution

In this paper, we propose a novel framework for the segmentation of noisy images by incorporating the advantages of k-means clustering and distance regularized level set evolution (DRLSE). Level set methods and active contour models (ACMs) plays a vital role in the applications of image processing, robot vision, object recognition and computer vision. DRLSE model has recently become a powerful technique for image segmentation. DRLSE model eliminates the reinitialization problem in conventional level set method. DRLSE has been applied successfully into some fields like medical imaging, remote sensing and computer vision. However when it is applied to noisy images, It leads to significant drawbacks (number of iterations and computational time is increased). In order to avoid disadvantages of conventional DRLSE, we introduce a method to combine the median filtering, k-means clustering and DRLSE model. Firstly a noise free image is extracted by median filtering; then k-means clustering is applied to denoised image. The last stage is that the DRLSE model is applied for the extraction of object boundaries with pre segmentation process. The accuracy and the efficiency of the algorithm can be described on various noisy magnetic resonance (MR) brain images. The experiments show that our proposed method is more effective for noisy images.