Repository containing work applying traditional statistical approaches on foreground segmentation for a Cheetah.

Techniques applied range from Bayesian Decision Theory with a Naive Bayes classification approach, Maximum Likelihood Estimation, Maximum A Posteriori Estimation to more complex models such as using Expectation-Maximization algorithm to Gaussian Mixture Models with different comparitive analysis by varying different hyper parameters such as cluster size and different subsets of the training dataset.

Note: Feature extraction was done using Discrete Cosine Transform on 8x8 blocks of the test image to generate a 64-Dimensional vector.