Expectation-maximization algorithm (EM algorithm) is an unsupervised learning algorithm for discovering latent variables from observed data. Image segmentation is an image processing procedure to label pixels of similar kind into the same cluster groups. A number of literatures has been investigating the possibility of applying EM algorithm on image segmentation. In this project, we experiment the performance of EM algorithm on image segmentation in comparison with K-Means algorithm as a benchmark.
Image Segmentation, K-Means Algorithm, Expectation-maximization Algorithm, Clustering
We applied EM algorithm on image segmentation and compare the result with K-Means algorithm. We consider both gray-scale images and RGB images. On top of that, we try to optimize the performance of image segmentation with some image preprocessing steps and a slight modification on the EM algorithm. The optimization procedures would be elaborated in the later session. In this session, we would generally introduce general EM algorithm, Gaussian Mixture Model and the problem formulation.