【作者】 曲若雷； 马艳梅； 潘地林；

【Author】 Qu Ruolei, Ma Yanmei, Pan Dilin(Computer Department in Anhui University of Science and Technology,Huainan 232001, China)

【机构】 安徽理工大学计算机系；

【摘要】 在基于视频监测的人脸识别中,由于人脸姿态随时呈现出的多样性,使得形态轮廓线同时产生复杂的变化,相应地,区域内的关键特征也会呈现很强的不稳定性甚至间断性,与原始几何形态有着显著的畸变。本文将论述如何在不稳定的实时图像序列中,通过性能监测方法和图像坐标上的仿三维旋转校准,获取最大正确率的关键特征测度;对后台图像数据库由内存中的单变量分层分区索引设想,到多维特征间的相互约束,最终可用各种先进方法做最后的目标匹配。阈值性能链的建立,就是为了保证预处理时特征来源的最优,这会充分运用视频的丰富资源,导致极好的识别能力,这是本文一直强调的重点。最后结合VC++ .Net和Canada Coreco Imaging机器视觉平台完成本识别方法的试验。更多还原

【Abstract】 Face recognition from a video is a capturing process from complex facial 2D-topography due to the partial or flexible performance measure human faces give to, thus identities probably break as time proceeds. This context systematically evaluates how identities of time-variant probe series be executed in real-time 3D registration model, and how the determined identities chase to background gallery, shrinking record area via one-dimensional cascading indices first and then combining to perform multidimensional restraint focuses, finally which method be the best for final validation. Establishment of first-order performance list with a threshold makes the inter-probe identity source optimally exploited. Testing research is based on VC++.net and computer vision camera interface from Canadian Coreco Imaging co, ltd.更多还原