一种新的图像中人眼定位方法

摘要

The human eyes are usually directly located in the target images,so it is difficult to ensure higher accuracy and difficult to achieve it. This paper uses the "human face-human eye" structure. Firstly,a skin color model in YCbCr space is used for roughly locating regions of probable face. Then the results of the training of support vector machine classifier composed of the human eyes are used for prime positioning. Finally, it completes the windows merge. Experimental results indicate that the algorithm largely improves the locating speed and accuracy with good robustness.%采用一种"人脸-人眼"两层结构的定位方法.首先利用YCbCr空间的肤色模型粗定位可能的人脸区域，然后利用支持向量机的训练结果组成分类器，对人眼进行初检，最后根据人脸特征完成信息融合，最终标定人眼.实验结果表明：该方法在简单及复杂背景下都能够得到较高的定位速度和精确度，具有较好的鲁棒性.