人脸轮廓定位和归一化的研究

研究了准确定位出从头顶到脖子的人脸曲线外轮廓的算法。采用人脸粗检、边缘检测、区域连通性滤波等方法。其思路是，首先粗检出人脸大致的矩形外轮廓，再做边缘检测、有效信息二值化、区域连通性滤波、轮廓点修正、对优先区域和整幅图像做垂直水平投影等处理，获取准确的人脸轮廓。还研究了通过比例变换、直方图修正等获得归一化面像。本算法适用于精确定位单一背景的人脸轮廓，复杂背景下也有较好效果。实验结果表明了本算法的有效性。本算法已经成功地应用于一个面像识别系统。

Facial Curvilinear Lineament Extraction and Unifying

The methods of face rough extracting, edging detecting and section connectivity filtering are presented to extract facial curvilinear lineament. The main processes of the algorithm are that, firstly, rough rectangular lineament of face is detected by rough face extracting, then some methods are taken orderly, which include edging detecting, effective information dimidiating, section connectivity filtering, revising of the contour points, vertical and horizontal projection of preferential section and whole image, and finally, the borders of facial image are delimited to dispose of the disturbing of background. The unifying of facial image is discussed later. The accurate delimiting of facial lineament is aiming at the simplex background, as well as the complex. The experiment results confirm that the algorithms are efficient. The algorithms have been successfully applied in a face recognizing system.