【摘要】 图像中不同亮度和颜色的光照是影响人脸检测的重要因素。本文提出一种新的肤色模型(H-SV-V肤色模型),它能有效地克服由于光照的亮度变化给检测造成的困难,检测出各种亮度条件下的肤色像素。对于有有色光照射的图像,我们对它进行色彩平衡,可以消除可能的色光干扰;在色彩平衡之前,先判断是否有有色光照射,这是通过分析图像中高亮区的特性来进行的。对于用该肤色模型提取出的肤色区域,我们是通过两个灵活的数学形态学算子来处理的,其一是目标提取算子,用来获取人脸候选目标(并采用“网格搜索”来加快获取速度);其二是关联算子,用来确认人脸候选目标。这一系列方法使我们能有效地克服光照影响,准确地检测出图像中不同大小,各种偏转方向,有大面积遮挡,以及有阴影的人脸。同时我们的检测速度也是较快的,精心设计的大量实验的结果也证明了这一点。更多还原

【Abstract】 Light condition of an image in varying brightness and color is an important factor affecting face detection. In this paper we bring forward a new skin model(H-SV-V Skin Model), which can effectively smooth away the difficulty brought from variety of brightness and detect skin pixels in varying light condition. For those ’images in colorful light, we carry out an operation named color-balance, which can eliminate disturbance of colorful light. But it should make a judge if there is really colorful light in image before carry out that operation. We can make that judge through analyzing character of high-lightness area. We use two morphological operators to process those skin areas extracted using our skin model. One is object-extracting operator which can be used to extract objects of face candidate(We also quicken this progress by ’grid-search’). The other is associating operator which can be used to affirm those face candidates. Through that series of approaches, we can effectively smooth away the difficulty of light condition, and exactly detect faces in image for any size, kinds of orientations, big-size occlusion or existing shadow. At the same time, our processing speed is higher than that of predecessors, which can be made know by results of abundant handpicked experiments.更多还原