Efficient and Robust Face Detection from Color Images

The efficient detection of human faces from images is fundamental in a variety of applications. Although various approaches to face detection have been developed, it still remains a challenging and difficult problem due to internal factors and external factors such as scale, lightning conditions, contrast between face and background, face regions overlapped one another. Especially when two or more faces are close to each other, their corresponding skin areas in the segmented binary image will be connected by ears or high luminance hairs. A significant challenge is the separation of joined faces from each other. The main method used for this separation process is erosion operation for the binary images at present. This method, however, is highly sensitive to the shape and size of the used structuring element. A large structuring element will entirely erase some binary regions, whereas a small element will have little effect at all. Noticing that there is in general more horizontal connectedness between faces, a simple and efficient method to segment overlapped facial regions is presented in this paper. Firstly, we extract facial regions based on skin information. Secondly, vertical integral projection method is adopted to segment the connected region with faces overlapped horizontally into several separate regions. Finally, the separate region is determined whether it is a face or not by integration of edge information, morphological analysis and template matching. Experimental results have demonstrated that the proposed algorithm is robust against color images with face overlapped horizontally and different types of skin with a complex background