不变区域提取算法及其在两视图粗校正中的应用  

An Extracting Algorithm of Invariant Regions and Application in Rough Rectifying for Two-View Images

作　　者：李竹林[1] 赵宗涛[1] 綦海龙[1] 翟晓颖[1] 

机构地区：[1]第二炮兵工程学院

出　　处：《微电子学与计算机》2007年第8期92-94,共3页Microelectronics & Computer

基　　金：二炮科研基金项目(EP2004-062)

摘　　要：在宽基线成像过程中,两视图的成像角度、光照条件都发生较大的变化,使得立体匹配的难度加大。提出了一种基于不变区域提取算法,通过灰度指数变换,定义的目标函数不仅提取了定位准确的极值点,形成稳定的不变区域,而且突显了每个方向上灰度的变化。将其应用于两视图RTS变化参数估算中,正确率分别达到:旋转91.65%,水平位移84.85%,垂直位移80%,比例84.29%。根据估算值对右图像进行校正,校正后的右图像明显地提高了与左图像的相似度。In wide baseline imaging process, viewpoint and illuminance are all changed more for two-view images, it makes stereo matching difficult. An extracting algorithm with invariant regions based on exponent transform was proposed. Defined object function can extract the extremal points located accurately, form the stable extremal regions, and magnify the local intensity changes. Using it to estimate RTS values, right rates of rotation, horizontal displacement, vertical displacement and scale are 91.65%, 84.85%, 80%, 84.29% respectively. The similarity between the right image rectified and the left image is improved obviously.

关 键 词：指数变换 极值点 不变区域 粗校正 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]