适应内容的广角图像拉伸畸变自动校正算法  

Automatical content-adapting stretched distortion correction algorithm for wide-angle images

作　　者：杨波[1] 冯华君[1] 徐之海[1] 李奇[1] 陈阔[1] 

机构地区：[1]浙江大学现代光学仪器国家重点实验室,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2016年第9期1711-1717,共7页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(61475135)

摘　　要：针对广角图像中视场边缘物体的拉伸畸变,提出一种基于图像内容的混合校正方法.利用线性透视投影的优点以及Mercator投影的保形性,通过有效组合线性透视投影和Mercator投影,平衡图像背景不弯曲和前景物体不被拉伸这一不可调和的矛盾.基于人脸检测算法,构造适应图像内容的自适应校正系数,并保持背景内容不被弯曲且前景物体不被拉伸.该算法不需要任何人工干预,可以实现自动化处理.实验结果表明,所提出的算法优于传统的投影校正算法,校正后的图像视觉效果良好.A hybrid content-based image correction method was presented for stretched distortion of objects near the edges of wide-angle images’ filed of view. With the advantage of linear perspective projection and the conformality of Mercator projection, the contradiction was balanced that neither bend straight lines in the background nor locally distort the shapes of foreground objects through the effective combination of linear perspective projection and Mercator projection. Furthermore, correction coefficient that adjusted to image content automatically was generated through face detection algorithm to keep lines in the background of images straight and to maintain the foreground objects not being stretched. Without any human inter-vention, our algorithm can automatically process distortion images. Simulation results show that the algo-rithm is convenient and practical; the results of correction is visually well.

关 键 词：拉伸畸变 自动校正 广角图像 线性透视投影 Mercator投影 

分 类 号：TP37[自动化与计算机技术—计算机系统结构]