A system is presented that is capable of automatically detecting specific locations of major facial features such as the eyebrows, eyes, nose, mouth, and facial boundary in portrait images. The system uses multiple stages to determine the feature positions. These stages include skin segmentation, eye detection, and facial feature finding. Each successive operation uses the output of the preceding step. Efficiency is achieved by using the least expensive operations to constrain the search area for the more expensive ones. A robust system has been created wherein each stage is tolerant of inaccuracies in the preceding stages.
展开▼