在当前主流的基于统计学习和穷举搜索窗口的人脸检测技术中,使用分类器对每个候选窗口进行分类判决之后,会留下多个重合的人脸候选窗口。为了将多个重合窗口合并为一个,提出一种使用矩形环对一个真实人脸的多个具有重合性质的分类器响应候选窗口进行重合范围限定,并合并多个窗口的策略。提出的矩形环合并窗框方法具有简单、易编程的特点,实验表明了该文方法的有效性。
In the most popular face detection methods which are based on statistical learning and brute-in-force search, after the search of each candidate sub-window, the surviving sub-windows for one real face are always overlapped each other. A rectangle loop is proposed to limit the overlap extent and used for removing multiple overlapped sub-windows. The proposed method is simple and easy for programming. The experimental results show its efficiency.