Face Features Localization Studies

Facial features localization is one of the key factors in the field of facial recognition system, facial cartoon recreation and facial tracing etc. Among main techniques of human facial features localization, the geometry shape information method has the high accuracy rate and the big robustness.This paper studies three geometry information based algorithms- Active contour model (Snake), active shape model (ASM) and active apparent model (AAM), and their application in human facial features localization.Consider characters of the human facial shape, the classical Snake model is modified by adding the shape energy item in its energy function in order to limit the illegal shape which appears in the iterative process. Experiments of human facial contour extracting indicate that the improved model performs better.After studying theories of the Active Shape Model and the Active Appearance Model, and their application in facial features localization, solutions to the model initialization and the illegal shape problem are explored. Firstly, eyes location is used to improve the precise of the model initialization. Then, the local shape constraint is used in the searching process for ASM, and the shape parameters for AAM's searching process is restricted and modified, which efficiently solves the problem of illegal shape. Experiments indicate that the revised strategy improves precision of facial features localization.Both ASM and AAM model need to label the feature points of many sample images, and the labeling accuracy directly affects the effectiveness of the models. In order to increase the accuracy and effectiveness of facial feature point labeling in training images, the paper designs a semi-automatic feature points labeling system.