一种基于SVM的网络不良图像过滤方法

摘要

针对网络不良图像过滤的需求,提出一种基于SVM 的不良图片快速过滤方法.该方法利用混合肤色模型实现裸露肤色区域的检测,提取人脸位置、形状和图像背景等特征,组成特征向量.用SVM分类器训练得到检测模型,利用这个模型进行判决,有效提高了不良图片的平均识别率.选取实际网络应用中的正常图像与不良图像,其中不良图像的识别率为83.9％,正常图像的识别率为93.4％,误检率为6.6％,平均识别率达到86.6％,实验显示该方法满足实际应用的需求.%For the demand of filtering pom image on internet, a new fast porn image filtering method based on SVM is presented. The method makes use of the mixed skin colour model to implement the detection of naked skin area, extracts the features of human beings face position, shape and image background, etc. To form the eigenvector. Then the SVM classifier is employed to train to obtain the detection model, and the model is then used in discrimination. The method effectively improves the average recognition rate on porn images. We choose normal and porn images from real internet environment for test, the recognition rate of pom image achieves 83. 9% , and the recognition rate of normal image achieves 93.4% , the average recognition rate achieves 86.6%. Experiment demonstrates that the method meets the demand of practical application.