Detecting Facial Features on Images with Multiple Faces

This paper presents an approach to detect facial features of multiple faces on complex background and with variable poses, illumination conditions, expressions, ages, image sizes, etc. First, the skin parts of the input image are extracted by color segmentation. Then, the candidate face regions are estimated by grouping the skin parts. Within each candidate face region, an attempt is made to find the eye pair using both Hough Transform and the Principal Component Analysis (PCA) method. If the predicted eye pair is, under the measurement of correlation, close enough to its projection on the eigen eyes space, the corresponding region is confirmed to be a face region. Finally, other facial features, including mouth corners, nose tips and nose holes are detected based on the integral projection algorithm and the average anthropologic measurements of the valid faces.