利用图像颜色及其边缘直方图特征的SVM人脸检测

研究了利用颜色直方图、颜色边缘幅值和边缘方向直方图特征,基于支撑向量分类器的检测人脸技术.提出了一种新的边缘方向编码,在与颜色直方图结合中比传统方向编码有更好的分类性能.采用多重交叉检验的ROC评估,实验表明结合颜色直方图和颜色边缘直方图能明显提高分类精度;支撑向量机能有效地挖掘基于颜色的综合直方图的分类潜力,分辨出不同光照条件下、不同表情甚至部分遮挡的非深度旋转的彩色人脸,体现出结合颜色及其边缘统计特征在人脸检测中的有效性和鲁棒性.

Using Color and Color Edge Histograms for SVM Based Human Face Detection

This paper studies the human face detection by support vector classifier using histograms of color, color edge and its orientation. A novel edge orientation coding is proposed, it outperforms the classification accuracy of the traditional edge orien- tation coding when they are combined respectively with color histogram. Using ROC evaluation in multi-fold cross validation, experiments shows that the classification accuracy can be significantly improved when the color histogram is combined with the color edge histograms. SVM can make use of the color based histograms for classification effectively, and can detect non-deep rotated human face in color image under different illumination, with different expressions and partial occlusion, which shows that the combination of color and its edge features is effective and robust for face detection.