一种改进的敏感图像过滤方法

针对现有敏感图像过滤方法误检率较高的问题，提出一种结合肤色检测和方向梯度直方图(HOG)人体检测的敏感图像过滤方法。采用HOG特征提取人体目标的特征集，运用支持向量机训练人体检测模型，检验图像中是否存在人体，并结合肤色检测算法判别该图像是否为敏感图像。实验结果表明，该方法能有效检测复杂背景条件下的敏感图像，其精确度为90.2%、查全率为86.3%、误检率为3.5%。

Improved Sensitive Image Filtering Method

Aiming at the shortage that existing sensitive image filtering method has higher error rate,this paper presents a sensitive image filtering method by combining skin color detection with human detection by Histogram of Gradient（HOG）.Extracting features of human bodies by HOG feature,using a detection model trained by Support Vector Machine（SVM） to find out the human body,and then using the skin color detection algorithm to tell whether the image is sensitive.Experimental results show that this method can detect the sensitive images under complex background effectively.The accurate rate can achieve 90.2%,the recall rate can achieve 86.3%,and the error rate can achieve 3.5%.