Face Detection Based on Segmented Multi-Gaussian and Radial Template
BAO Jun,GONG Sheng-rong,LIU Chun-ping,ZHOU Fei-fei
DOI: https://doi.org/10.19304/j.cnki.issn1000-7180.2009.10.008
2009-01-01
Abstract:In the current methods of face detection,the problem of the accuracy of face detection is relatively low and the error of rotation angle is relatively high in the plane,presenting a new method of combining the skin color regions,organ regions and Radian Template Histogram.Firstly,this method adopts improved Segmented Multi-Gaussian to partition skin color regions.Secondly,obtaining the centers of mouth and eyes through combining the region similarity,the threshold of skin color regions and geometric properties of facial organ in the skin regions.Lastly,using the histogram of the Radian Template,Obtaining three peaks and getting the three distances between them,Locating faces and their rotating angle.The results show that this method can locate the face rapidly within the error of 10°,without moving the center of Radian Template or change the radius.