Study For Face Detection Based On Panoramic Image Mosaics In Complex Background

Face detection firstly applied in face recognition as face positioning. Itgenerally can be described as: given a static or dynamic image, judge whether it hasfaces; if has faces, all the faces will segmented from the image, and determine thelocation and size of each face in the image. With face rec ognition applicationbackground is gradually complicated, face detection has developed into anindependent discipline, and gained the world’s attention. In face recognition, visualretrieval, intelligent monitoring, security, access control and other fields has a veryimportant application value.In this paper, based on recent participation in the project "recognition systembased on the video of classroom teaching and examination staff "to conduct research, mainly related to the knowledge of the image mosaic and face detection. This paper focuses on the complex background face detection algorithm based on the panoramic image stitching. The actual needs of the project on the image stitching technology, the panoramic image mosaic algorithm based on SIFT feature matching. Right, two and above a perspective view of a single panorama image by image matching technology and image stitching technology.Face detection algorithm for the requirements of the project, the focus of research. First, we propose a preprocessing algorithm of face detection based on skin color segmentation. Of skin color clustering characteristics, skin color clustering results comparison of different color space, we choose YCbCr space. Then, in order to improve skin tone and hair color clustering effect, the use of an adaptive illumination compensation method. And through the establishment of a Gaussian skin color model, skin color similarity calculation and optimal threshold segmentation algorithm skin tone and hair color segmentation, and finally by the mathematical morphology, excluding some non-skin color and hair color area. Experimental results show that the proposed face detection pretreatment method is feasible and effective. Collection of images in this article are complex background color face image, there is a lot of noise points, split out skin tone and hair color area and can not be directly and accurately locate the face area. Based on the situation, this paper presents acombination of geometric constraints and face mask facial hair face detection algorithm. Geometric positional relationship of the hair by first face and the face of rough detection, and acquires the face selected region, then the selected area of each f ace using improved final accurate face detection algorithm based on face mask face location, the experiment proved that the method for complex background color face image can be quickly and accurately detects human face.