【作者】 李烨；

【导师】 滕国库；

【作者基本信息】 大连海事大学 ， 计算机技术， 2011， 硕士

【摘要】 近年来通过互联网和移动通信网络传播的不良信息(主要是淫秽色情内容)对广大未成年人的身心健康造成了恶劣影响。这一问题日益引起社会各界的关注。目前不良图片和不良网站的过滤技术主要有三种：基于URL的过滤、基于文本内容的过滤和基于图像内容的过滤。基于图像内容的过滤技术相比于前两种技术具有更广泛的适应性和更高的灵活性。本课题着重研究基于图像内容的过滤。截至2010年9月,本人在中国知网CNKI上没有检索到国内有论文研究应用于手机浏览器的基于图像内容的不良图片过滤,在搜索引擎上也没有搜索到国内有已而市或推广的应用于手机浏览器的基于图像内容的不良图片过滤软件。本人选定基于Windows Mobile 6操作系统和ARM微处理器的智能手机作为开发平台,OpenCV作为图像检测模块的开发工具。经过对智能手机平台软硬件特点的分析,确定合适的基于图像内容的不良图片检测方法。本文提出了一种基于Winsock分层服务提供者LSP技术的浏览器色情图片过滤软件框架。由于Windows Mobile应用程序开发与桌面Windows应用程序开发有许多相似之处,而且OpenCV必须经过移植才能在Windows Mobile上使用,在开发过程中,为了方便调试和性能比较,首先实现了应用于桌面Windows的浏览器不良图片过滤软件,该软件的网络数据截获模块采用了LSP技术,图像检测模块采用了人脸检测、白平衡矫正皮肤检测、支持向量机SVM等技术(皮肤检测又包括肤色检测、纹理检测、皮肤连通区域检测等)。经验证,该软件在桌面Windows上可实时过滤浏览器访问到的不良图片,并具有较高的准确率。然后将原本不能适用于Windows Mobile的OpenCV库移植到了Windows Mobile 6平台上,并将已开发的适用于桌面Windows的浏览器不良图片过滤软件移植到了Windows Mobile 6平台上。经过测试,该软件在Windows Mobile手机上能正常工作,但由于智能手机的硬件条件限制,对于一般的智能手机而言,人脸检测的计算量较大,图像检测模块的耗时较长。更多还原

【Abstract】 In recent years, with the popularity of the Internet and mobile phones day by day, harmful information, especially pornographic information, spreads through the Internet and the mobile network, resulting in very negative impact on the physical and mental health of minors. The problem arouses more and more concern of society. Up to now there are three major methods for filtering pornographic images and pornographic Web sites, which are filter based on URL that is short for Uniform Resource Locator, filter based on text and filter based on image content. Compared to the first two methods, the adaptability of filter based on image content is wider and its flexibility is higher. The thesis focuses on content-based image filter.Up to September 2010, I found that there were no articles on content-based pornographic image filter for mobile phones in CNKI that is short for China National Knowledge Infrastructure and no software for mobile phones to filtrate pornographic images comes into market in China in search engines.Smart mobile phone based on Windows Mobile 6 operating system and ARM microprocessor is selected as the development platform, and OpenCV which is an open source computer vision library is selected as the development tool of the image-detecting module. After the features of the hardware and software of the smart mobile phone platform are analyzed, appropriate content-based methods for filtering pornographic images are selected. A software framework for filtering pornographic Web images is presented, which is based on LSP that is short for Layered Service Provider. Because the application development of Windows Mobile is similar to that of Windows desktop and OpenCV cannot be directly used on Windows Mobile, software filtering pornographic Web images for the Windows desktop is developed at first in order to facilitate debugging and comparision. The principle of the module hooking network messages is LSP, and the principle of the module on image-detectinging is face detection, white-balance correction, skin detection and SVM that is short for Support Vector Machines. And skin detection includes skin color detection, texture detetion and connected skin region detection. When it is tested on Windows desktop, the software can real-time filter pornographic images on webpages and its accuracy is fairly high. And then OpenCV and the software are both ported to Windows Mobile. Testing result shows that the software can work on a Windows Mobile phone, but because of the hardware limit of average smart mobile phones, the calculation amount of face detection is quite large, and the module on image-detecting is a litte slow.更多还原