图片卫士:一个自动成人图像识别系统

设计并实现了一个自动识别成人图像识别系统“图片卫士”。图片卫士采用3层识别框架，利用肤色、纹理、图像视觉特征分层逐级识别成人图像。为了可靠地检测到图像中的肤色区域，提出了一种新的自适应统计肤色模型。在肤色检测基础上，通过皮肤纹理验证过程，图像中的人体皮肤区域被准确地分割出来。基于图像中皮肤区域，提取9个经验特征来表示图像内容，并采用AdaBoost算法构造一个总体分类器进行图像分类，识别正常图像和成人图像。在算法评估中，建立了一个78205幅图像的测试集，其中59885幅为正常图像，18320幅为成人图像。图片卫士显示了良好的系统性能，具有成人图像88．5％的识别率，正常图像92．5％的识别率。在PentiumⅣ1．5GHz的个人计算机上，图片卫士的平均处理速度为正常图像每秒5．6幅和成人图像每秒1．9幅。图片卫士可以应用在个人计算机或网络传输中，实时监控和过滤成人图像，还可以为网络安全等应用提供技术支持。

Image guard: an automatic adult image recognition system

This paper proposes an image recognition system, called Image Guarder, to recognize adult images automatically. Image Guarder employs a hierarchical recognition framework with 3 layers, which integrates skin color detection, skin texture validation and image classification. A novel adaptive statistical skin color model is designed to reliably detect skin pixels. Based on skin color detection, skin regions are extracted by a smooth texture validation process. From the detected skin regions, nine ad-hoc image features are extracted to represent the image content. At last, the AdaBoost algorithm is adopted to construct an integrated classifier for the final image classification. To evaluate the performance of Image Guarder, a large image dataset consisting of 59 885 benign images and 18 320 adult images is set up. The proposed system shows an exiting performance that the precision for the adult images is 88.5%, and the precision for the benign images is 92.5% respectively. On the Pentium IV 1.5GHz PC, the average processing speed is 5.6 frames per second for the benign images, while 1.9 frames for the adult images. Image Guarder can be used to inspect or filter adult images on the personal computer or in the network transmission. It also provides a technique that supports for the network security applications.