Segmentation, grouping and feature detection for face image analysis

A system capable of analyzing face images in a wide range of orientations, such as ± 90° of turn or small to moderate tilt, is discussed. For images of size 256×384, the system can find the face orientation and facial feature groups within 40s on a Sparc10. Low-level analysis consumes about 30s. The overall analysis could be optimized to within 25s. Moreover, the discussion shows that eyeglasses constitute a significant class of objects requiring the same research effort as eyes, mouth, and nose. Skin regions are also important 'macro features' that contribute to the robustness of the analysis. The study also shows that segmentation, grouping, and feature detection techniques can be applied to other image analysis problem domains.