Face and Facial Feature Localization Based on Color Segmentation and Symmetry Transform

An effective approach for face localization, based on skin color segmentation and gray image symmetry transform, is proposed in this paper. To reduce the influence of background and variable facial poses, the Blob analysis and ellipse fitting methods are employed to process the image blobs detected by conventional skin color segmenting strategy. Invalid blobs are removed and the candidate face regions are adjusted vertically. Furthermore, segmenting hue image is adopted to extract mouth region from each candidate face region. Afterwards, the Generalized Symmetry Transform (GST) of gray image is used to obtain the candidate´s eyes and the real ones are detected by combined optimization strategy of the proposed cost function which is related to the model of facial geometry and positions of eyes and mouth. Finally, other facial features such as chin and cheek are then precisely located. The test results verify the robustness of the proposed method.