Research On Face Detection And Pupil Position

In recent years, human face identified system is required for applications to adapt the more general environment in the national public security and information security departments. So it requires the researchers to propose a more robust, efficient, real-time human face detection algorithm. Under the demand of this practical application, the research content of this paper including a human face region is divided up in the background image, the position of the pupil region and eye brow region are detected in it. The main research contents are as follows:1. Human face detection is the preliminary work of facial feature points detected and the location. People face has relative stability because skin color is the important information of human face and it is not dependent on the details of facial features, applicable to the rotation and the changes in expression. It also distinguished the color from most of the background objects. This paper adopts a method of combination of human face color segmentation and its geometric characteristics. Contrast to the analysis of color space and color model, the thesis choose the YCrCb color space and make use of the Gaussian model of YCrCb color space to do skin color segmentation. According to making analysis of the divided human face region, the thesis makes use of geometric characteristics of the region to locate the region of human face.2. This paper proposes a new method for eyebrow detection in the images on account of the traditional method of eyebrow detection vulnerable to Liu block and head tilt and other factors. With this method, we calculate the gray gradient absolute value of each pixel in the image on the eight domains, and choose the maximum value of the eight gray gradient absolute values as the pixel value of the pixel, then we use a sub-matrix quantify these complicated value in the processed gradient matrix. Then we choose the relatively large value blocks of the post-quantitative block, it has some regional overlap among these sub-blocks. Finally, we combined these gray sub-blocks to obtain the largest available external rectangular as the region of eyebrow.3. This paper proposes a new method for pupil position because the traditional methods of human eye position susceptible to the impact of scale, divided face image, then detect the extreme in the multi-scale space image, then remove the low contrast characteristics and fringe response point in order to precisely locate the position of the pupil. Because this method extract the part feature of images, it will maintain invariable to the rotate, zoom scale, changes in brightness, it also keep a certain degree of stability to noise. The experiments of this paper validate its advantages.