Extraction of face objects using skin color information
Young-Bum Sun, Jin-Tae Kim, Won-Hyung Lee · 2003
This paper attempts to suggest an efficient method to extract only the face region from the complex background as preprocessing to face recognition. Under this method, only the skin color is extracted by applying multi-channel skin color models not affected by brightness, and then the method of object extraction is applied so that multiple faces can be extracted from a single input image, distinguished from each other. This method solves the problems that the existing MER (minimum enclosing rectangle) extraction method has. Finally, candidate objects where faces are likely to exist are selected by checking the sizes of extracted objects, the XY ratio, and the distribution ratio of skin colors. The regions where faces are most likely to exist can be extracted just with the information of skin color, through application of multi-channel skin color models and object extraction, and based on the sizes of the objects, the XY shape ratio, and the skin color distribution ratio. Our method can increase the ratio of successful extraction in various environmental variables such as face extraction with complex background, slanted faces, and face with accessories.