基于SVM的血腥图片检测方法  

A SVM-Based Bloody Image Identification Method

作　　者：魏远旺[1] 殷海明[1] 

机构地区：[1]嘉兴学院数理与信息工程学院,浙江嘉兴314001

出　　处：《嘉兴学院学报》2013年第6期121-126,共6页Journal of Jiaxing University

摘　　要：通过一个快速的血色模型,标出图片中的血色区域,然后对血色区域进行特征分析,提取最大块的周长-面积分形维数表示其形状特征,提取最大块的PIM图像信息熵表示其纹理均细度.同时,提取图片中的肤色信息进行辅助判定.最后,把得到的由颜色、形状、大小、肤色、纹理等特征构成的七维矢量输入到一个经过训练的SVM分类器中进行判断.通过对互联网上随机下载图片的测试表明,该方法对血腥图片有较高的检出率,而对常见的各类正常图片误检率较低.The blood regions in the image are identified by a quick blood model, and then the perimeter-- area fractal dimension of the biggest region is calculated to describe the shape; the PIM image information entropy of the biggest region is calculated to represent the texture. The skin information of the image is also ex- tracted as a supplement for the judgment. At last, the feature vector consisting of color, shape, area, skin in- formation and PIM is fed to a trained SVM--classifier. The experiments on images downloaded randomly from the internet indicate that this method can detect the bloody images with high recall ratio and low false positive ratio.

关 键 词：血色模型 分形维数 图像信息熵 肤色检测 Support Vector Machine(SVM) 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]