基于内容的敏感图像特殊部位识别算法研究-应用数学专业论文.docx
文本预览下载声明
I
I
摘 要
摘 要
随着宽带网络的普及与发展,网络信息已成为人们生活的一部分,给人们的生活 带来了极大的便利,但同时也带来了许多淫秽不良信息,严重影响了人们的日常生活。 为此,如何过滤网上不良信息,营造一个健康的网络环境,成为人们关注的焦点。
本文主要研究工作如下:首先,对肤色检测的相关技术进行了详细陈述,研究了 YCgCr、HSV、YCbCr 三种不同的颜色空间,通过比较分析,选择了肤色分类效果更好 的 YCgCr 空间作为肤色检测的颜色空间,然后在 YCgCr 上建立高斯肤色模型进行肤色 检测,并利用加权均值滤波对图像进行平滑去噪,实验表明,该方法能有效地检测出 肤色区域。其次,由于肤色模型具有一定的局限性,会把一些颜色比较接近人体皮肤 的其他区域误判为肤色区域。为此,加入人脸检测的方法进一步过滤排除,本文比较 了两种人脸检测算法:基于 BP 神经网络算法和基于 Adaboost 算法,通过比较分析得 出在分类性能上 Adaboost 算法优于 BP 神经网络算法。再次,对敏感图像特殊部位识 别算法框架进行了详细介绍,对于女性胸部识别本文采用了阈值法来进行胸部识别定 位;对于女性下半身特殊部位识别,本文提取了 Haar-like、肤色、HOG 三个特征, 并通过 Adaboost 算法训练得到了分类器,对该分类器的分类效果进行了比较。最后, 通过对过滤系统的实验分析和性能测试可知,本文所提出的方法能快速、有效的过滤 敏感图像,而且应用广泛,不但可以应用到固定的电脑平台上,还可以应用到移动的 手机平台。
关键词: 敏感图像;肤色检测;人脸检测;特殊部位;特征提取
Abs
Abstract
II
II
Abstract
With the popularization and development of wide band networks, the network information has become a part of people’s lives, and the network information has brought great convenience for people’s lives. But network information has also brought a lot of pornographic information, which has seriously affected pe1ople’s daily life. Therefore, how to filter online information for a healthy network environment has become the focus of attention.
This paper mainly research wok as follows. Firstly, the paper introduces the skin detection principle, the main research of this thesis includes YCgCr color space, HSV color space, and YCbCr color space. With the result of that the paper selects the YCgCr color space to build Gaussian skin model. The weighted average filter is used to eliminate images’ noises. The results of experiment indicate that the method can effectively detect the sin regions. Secondly, because the limitation of skin color model, so some more similar to human skin color other region always are recognized as skin area. Therefore, this article uses the face detection method to filter the color region. This paper compares the two types of face detection algorithm’s classification performance: base on BP neural network algorithm and base on Adaboost algo
显示全部