Segmentation of Noisy and Textured Images by an Expectation Maximization and Gradient Descent Iteration Method
Zhenzhou Wang, Yongming Yang · 2015
Segmentation of noisy and textured images remains challenging in both accuracy and computation efficiency. In this paper, we propose a new method for segmentation of noisy and textured images. The proposed method is based on the famous Expectation Maximization (EM) methodwhich calculates the global parameters of the image and Gibbs distribution which calculates the local parameters of the image. With the global parameters of the objects and the background computed from EM, a pre-segmentation is achieved. Then we propose a gradient descent iteration (GDI) method to achieve the final segmentation by minimizing the sum of local energy. Experimental results show that the proposed method is more effective than the state of art Normalized Cut method in segmenting noisy and textured images.