New robust image operators and applications in automatic facial feature analysis

This paper discusses a collection of image operators we originally developed for automatic analysis of face images, but they can also be applied to many other image domains. Most of these are new operators, a few are enhanced variants: two region segmentation algorithms (edge-based and intensity-based), two feature detectors (a hybrid morphological-Laplacian, and an oriented morphological edge detector), a thin edge detector by morphological gradient with statistical thresholding, and a nonlinear smoothing filter (micro-clustering). These new operators were designed with specific criteria for maximum efficiency in automatic image analysis. Morphological, statistical, linear and nonlinear operators were extensively tested and combined to get the desired properties.