Classifier based on Gaussian Mixture Models obtained from Expectation and Maximization Algorithm. In this code, the joint probability distribution in form of Gaussian Mixtures is computed for each dimension (features consisting of DCT coefficients) of each class (foreground and background). During each step of EM, points in the cheetah image get a soft assignment to each of the C gaussians. Next, having obtained the Gaussian Mixture Models, we use Bayesian Decision Rule to determine which points in the test image belong to the foreground and which of them belong to the background. 

Assumptions: Covariance matrices are diagonal because we’re learning GMMs for each of the dimensions separately