Pixel Classification and Heuristics for Facial Feature Localization

摘要

In his work, we use a broad set of pixel features of low computational cost-which includes first order gray-level parameters, second order textural features, moment invariant features, multi-scale features, and frequency domain features-for pixel classification based on facial feature localization. A Radial Basis Function Neural Network performs the classification into three regions of interest. Morphological filters and intrinsic geometric properties of the human face are combined into a post-processing heuristic to finish the feature localization. We present the results, which are qualitative and quantitative satisfactory.