【作者】 王志强；

【导师】 周敬利；

【作者基本信息】 华中科技大学 ， 计算机系统结构， 2004， 硕士

【摘要】 人脸检测是计算机视觉和模式识别领域中的一个热点问题,其不仅具有广泛的应用背景,也有极大的理论研究价值。一般而言,复杂背景、多人脸和变姿态的人脸检测难度较大,现阶段研究较多的是对图像中的人脸个数、位姿、角度、尺寸、偏向等进行了一定限制的人脸检测问题。在分析和实验的基础上,提出了一种基于彩色图像多线索信息的人脸检测算法。该算法综合利用了肤色分割、区域分析和统计分析等原理,以对输入图像进行人脸检测,并标示出(如果存在)每个人脸所在的位置。算法总体上分为三个步骤。第一步,利用提出的一种基于混合高斯分布和亮度自适应相结合的肤色分割算法,对输入的彩色图像中的肤色像素进行提取。该算法在提高了正检率的同时,也在一定程度上降低了错误报警率。另外,该步中还采用了亮度阀值分割方法,有效地去除了肤色分割中形成的部分类肤色像素,进一步缩小了待检区域的范围。第二步,根据人脸形状和结构的先验知识对得到的待检区域进行基于数学形态学和结构模式识别的分析和处理。该步骤的结果是生成了一系列的候选人脸对象。第三步,构造了一种基于最小风险的贝叶斯分类器,并利用其对候选人脸对象进行二分类判决,把属于人脸的对象标示出来。从而最终完成了人脸检测的任务。实验结果表明,这种多线索的彩色图像人脸检测算法具有较高的检测正确率和低的错误报警率,同时也具有较快的检测速度,在自适应能力和稳定性方面也达到了预期结果。更多还原

【Abstract】 Face detection is a hot problem in the fields of computer vision and pattern recognition, with the purpose of localizing and extracting the face region from the background. A human face is a dynamic object and has a high degree of variability in its appearance, which makes face detection a difficult problem in pattern recognition. However, with the further study in face detection, it will surely speed the development of artificial intelligence, computer vision, and other relative subjects. In this paper, a multi-clue based face detection algorithm for color image is presented. First, the image with human faces is input and dealt with a Mixture-of-Gaussian (for multimodal distributions) model and a self-adaptive luminance model to get all skin color regions, as is of great importance to fast face detection although some skin-like color regions are often chosen falsely in the mean time. As a result, a binary image with face regions is obtained. The binary image can be further divided into several sub-regions, which are then analyzed so as to eliminate useless non-face regions according to the facial features and some general knowledge. Consequently, all candidate face regions with special position and size are acquired. And each candidate region is required to include not less than two holes, because double eyes are generally darker than other regions in human face and thus often make holes in their positions. In the end, using Bayes decision, a probability model with the distribution of human face is constructed to perform the final face detection. Experimental results demonstrate the high probability of correct detection and adaptive abilities of the algorithm, and a decent speed of the face detection is achieved.更多还原