Visual Abstraction of Wildlife Footage Using Gaussian Mixture Models and the Minimum Description Length Criterion