As an alternative to the computationally intensive GMMs,
Lee and Yoo [63] proposed an “elliptical boundary model”,
whose performance is comparable to that of GMM and
yet the computational complexity is as simple as training a
SGM. The elliptical boundary model is defined as