Connected components labeling based face detection and pose estimation using PCA

Abstract

Based on a combination of color segmentation, connected component labeling. and morphology, an algorithm for human face tracking and principal component analysis based head tilt angle calculation is developed. The method uses HSV space for color segmentation since in this space the dynamic range of the skin color is quite narrow thus enabling us to differentiate human face from other objects in the scene. The connected component labeling, and, morphology help the segmentation process by removing noise artifacts in the scene and making the proposed algorithm robust to environmental conditions. The performance of the proposed system is evaluated for a large set of video sequences. It is found that the error in horizontal and vertical (x-, y-) translations, and the tilt angle is negligibly small for both studio and real environments.