采用动态HMM概率分布模型的人眼精确定位

从含有复杂背景的单幅灰度图像中精确定位人眼仍是一个尚未完全解决的复杂问题.为了对图像中的人眼进行精确定位,提出了一种基于动态HMM概率分布模型的人眼精确定位方法,该方法采用了含状态持续时间的动态观测符号概率分布HMM模型,首先用虹膜网格采样方式和特殊的特征提取办法来抽取观测序列;然后通过对观测序列进行评估来控制采样网格大小,并动态修正观测符号概率分布.这样无需对图像进行旋转、缩放和匹配运算,即可对图像中的人眼进行精确定位.实验结果表明,该方法检测效率较高、算法鲁棒,并具有较高的定位精度.

A Precise Eye Localization via a Dynamic Probability Distribution HMM Model

Accurately locating the human eyes in a single and gray level image with cluttered background remain a difficult problem in recent year.In this paper,we use a discrete Hidden Markov Model(HMM) that included explicit state duration density to locate the human eye precisely.At first,a retinal sampling grid and a special method are adopted to extract the observation sequence.We do not need rotation and scaling transformation and matching operation,for we use a dynamic observation symbol probability distribution in state to adapt the various angles of human eyes in image,and a dynamic sampling model that controls the sizes of the retinal sampling grid through the evaluation of the observation sequence to adapt to the various sizes of human eyes in image.The experiment results show that our algorithm is effective and robust and have high positioning accuracy.