【摘要】 皮肤识别是人脸识别、视频监控等技术的基础。提出了一种新的皮肤检测方法。首先,图像被划分成8×8的区域块,并从每个块中提取3个纹理特征,分别表达其水平、垂直和对角3个方向的纹理,经一个基于贝叶斯方法的纹理特征模型判断,识别皮肤块。然后,提取块中64个像素的两个颜色特征,采用正反查表法识别块中皮肤像素。该方法提取了皮肤的纹理和颜色特征,并在8×8块和像素两个尺度上综合了SPM和查表法的优点。实验结果表明该方法能够在精确识别皮肤的同时保持较低的背景错分率,取得了较好的效果。更多还原

【Abstract】 Skin detection is the basic technology of human face recognition, video monitoring, etc. A new skin detection method is presented. Firstly, an image is divided into 8×8 blocks from each of which 3 texture features can be extracted. These features are sensitive to the horizontal, vertical and diagonal textures accordingly, and they are utilized by a bayesian method based block texture feature model to distinguish the skin blocks. Then, two color features are extracted from 64 pixels in every block. Every pixel can be filtered using so-called positive-negative lookup table method. This skin detection method integrates skin color and texture features to detect skin on multi-scale with SPM and look-up table method. The experimental result shows that this method can maintain low mis-recognition error while high skin recognition rate can be reached.更多还原