| 摘要: |
| [目的]针对肤色检测易受光照影响的问题,提出一种基于边缘检测和灰度投影的人眼定位方法。[方法]首先结合肤色检测和Sobel边缘检测来提取人脸主要特征,得到人脸特征的二值化图像;其次根据人眼在人脸的几何位置关系得到人眼的粗定位;然后通过对数变换处理定位后的图片;最后进行水平和垂直方向的灰度投影,经过曲线拟合寻找极值进行人眼瞳孔的精确定位。同时,将本算法与其他类似算法进行比较分析。[结果]本算法对于不同光照和干扰环境有一定的适应性,对于不同姿势的人脸也能准确定位,相对于传统的肤色检测在精准度上有所提升,但在人脸图像受到环境或者噪声干扰严重时,本算法的定位成功率明显降低。[结论]该方法简单实用,对光照和复杂的干扰环境有一定的适应性,并且在一定的角度范围内,具有较高的准确性。 |
| 关键词: 人眼定位 边缘提取 几何提取 灰度投影 曲线拟合 |
| DOI：10.13656/j.cnki.gxkx.20170525.001 |
| 投稿时间：2017-01-23修订日期：2017-03-17 |
| 基金项目:国家自然科学基金项目(61401108),广西自然科学基金项目(2016GXNSFAA380154)和"大学生创新创业训练计划"广西壮族自治区立项项目(201610593187)资助。 |
|
| Eye Localization Method based on Edge Detection and Gray Projection |
|
ZHOU Liyu, CHANG Kan
|
| (School of Computer, Electronics and Information in Guangxi University, Nanning, Guangxi, 530004, China) |
| Abstract: |
| [Objective] In eye localization, human skin color is easily affected by light.To solve this problem well, this paper proposed an eye localization algorithm based on edge detection and gray projection.[Methods] Firstly, the proposed method extracted the main characteristics of human face by jointly using the Sobel edge detection method and the detection of the skin color to get the binary images of human faces.Secondly, the rough localization was obtained according to the geometrical positions of human faces. Afterwards, Logarithmic transforming was used to enhance the contrast of pupils and eyeballs. Finally, gray projection was performed in both horizontal direction and vertical direction, and the accurate position of the eye pupil was located by using curve fitting method to find the extreme value. Meanwhile, the algorithm was compared with other similar algorithms.[Results] The algorithm which adapts to different light intensity and complex interference environment can pinpoint the face with pose variation accurately. Compared with the traditional skin color detection, the accuracy is improved. But when the face image is disturbed seriously by the environment or the noise, the localization success rate of this algorithm is obviously reduced.[Conclusion] The experiment results showed that the proposed method could be adapted to different light intensity and complex interference environment. In a certain angle range, the proposed method achieved high accuracy. |
| Key words: eye localization edge extraction geometrical extraction gray projection curve fitting |
