Maximum Likelihood Estimation

Synonyms

Maximum likelihood estimator

Definition

Maximum likelihood estimation seeks to estimate model parameters that best explain some given, independent measurements according to a noise model.

Background

Many problems in computer vision can be formulated as finding the parameters of a predefined model given measurements or training examples.

For example in image segmentation one may want to describe a region by a simple region model, e.g., by a constant intensity value μ. There are many measurements, namely all the pixel intensities in the region. Assuming that these pixel intensities are independently generated from the constant intensity model according to a Gaussian distribution, the goal is to find the most likely parameter μ given these measurements. In this simple example, the optimal parameter μ is the mean of all intensities.

There are many more similar problems in computer vision, for instance, in the scope of optical flow estimation, camera calibration, image denoising,...