Abstract

A practical algorithm for face detection is presented to locate multiple faces in a color scenery image. An image is first classified by skin/non-skin algorithm. Then, morphology algorithm is used to clean the skin-likes image map by a pre-defined face element structure. Next, the image is clustered and region is generated according to processed pixels left in the image. After region separation and coarse region detection, we introduce an elliptic face mode by using eigen vector to locate a candidate human face. Finally, a face verification process is implemented to detect a real human face for all candidate face regions. A face center and pose is detected and modified by rotation. The main technique in facing location are color, shape detection and noise cleaning. In this implementation, there is no size limitation for color scenery image and human face inside. Experimental results cover different phases. Critical problems in face location still need to be explored. This project is implemented in Java.