Edge Preserving Image Segmentation using Spatially Constrained EM Algorithm

In this paper, a new method for edge preserving image segmentation based on the Gaussian Mixture Model (GMM) is presented. The standard GMM considers each pixel as independent and does not incorporate the spatial relationship among the neighboring pixels. Hence segmentation is highly sensitive to noise. Traditional smoothing filters average the noise, but fail to preserve the edges. In the proposed method, a bilateral filter which employs two filters - domain filter and range filter, is applied to the image for edge preserving smoothing. Secondly, in the Expectation Maximization algorithm used to estimate the parameters of GMM, the posterior probability is weighted with the Gaussian kernel to incorporate the spatial relationship among the neighboring pixels. Thirdly, as an outcome of the proposed method, edge detection is also done on images with noise. Experimental results obtained by applying the proposed method on synthetic images and simulated brain images demonstrate the improved robustness and effectiveness of the method.