摘要
目前已涌现出了许多人脸检测算法,而每种算法的侧重点不同,不能形成很好的综合检测能力。为了兼顾人脸检测的TP,FP和检测时间,文中提出了一种基于改进型LBP纹理特征的Floatboost算法。该算法首先提取具有一定旋转不变性的改进型LBP纹理特征。其次采用双阈值Floatboost算法来训练特征值,并生成强分类器。最后采用Adaboost级联算法来进一步减少检测时间。通过实验分析表名:该算法在保证一定TP的前提下,不仅可以减少检测的时间,还降低了FP。
There are numerous algorithms for face detection with different emphasis. Combining the truepositive（TP） rate, false-positive （FP）rate with the detection time, an improved Floatboost algorithm is presented based on the modified LBP texture features. The algorithm can extract the modified LBP features with certain direction invariance. And the FloatBoost with double thresholds is used to train the feature and generate strong classifier. Finally,the Adaboost cascade algorithm is implemented to reduce the detection time. Experiment results show that while the TP rate is guaranteed, the algorithm can reduce the detection time and lower the FP rate.
出处
《南京邮电大学学报（自然科学版）》
北大核心
2014年第6期75-79,共5页
Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition
基金
国家自然科学基金(61106021)
南京邮电大学校科研基金(NY211059)资助项目
