面向图像的垂直搜索引擎关键技术研究

发布时间：2018-03-15 20:55

本文选题：垂直搜索　切入点：图像内容检索　出处：《大连海事大学》2013年博士论文　论文类型：学位论文

【摘要】：随着网络带宽和软硬件性能的不断提高,多媒体文件的搜索需求日益高涨,图像检索作为文本检索到媒体检索跨域的第一步,逐渐成为是一个热门的研究领域。本文针对图像检索领域,从基于图像上下文的搜索和基于图像内容检索两方面进行研究。首先,研究了基于图像上下文垂直搜索引擎中的网络蜘蛛技术进行网络蜘蛛抓取和主题判定,使用DOM树和正则表达式对采集后的信息进行上下文提取,对提取内容切词建立索引,在满足互联网应用的同时,利用移动互联网技术,将内容转化至移动终端,并使用树状表示法建立规则知识库,实现结果的个性化推荐。然后,研究了基于内容的图像检索技术,将图像分层分割后,提取图像颜色、纹理和边缘特征进行多种特征融合,建立基于模糊支持向量机的多项分类模型,通过模型对图像不同图层的分类将多个关键字分配至该图像,使图像内容变成多关键字的描述最后,在图像检索的基础上,提取图像视觉内容特征中的纹理和颜色,抓住图像间的内容特点改进色彩传递方法。一方面,通过计算图像间的纹理相似度,搜索内容与灰度图像相近的色源图像,提高染色的成功率,另一方面提取颜色特征对图像进行更科学的像素采样,改进色彩传递算法本身,加强染色效果。
[Abstract]:With the improvement of network bandwidth and the performance of hardware and software, the search demand for multimedia files is increasing. Image retrieval is the first step of text retrieval to cross-domain media retrieval. This paper focuses on image retrieval from two aspects: image context-based search and image content retrieval. Firstly, the web spider technology based on image context vertical search engine is studied for web spider capture and theme determination, and DOM tree and regular expression are used to extract the context of the collected information. In order to meet the needs of Internet application, the content can be converted to mobile terminal by using mobile Internet technology, and the rule knowledge base can be established by using tree representation to realize the personalized recommendation of the result. Then, the content-based image retrieval technology is studied. After the image is segmented, the color, texture and edge features of the image are fused, and a multi-item classification model based on fuzzy support vector machine (FSVM) is established. Multiple keywords are assigned to the image by classifying the different layers of the image by the model, so that the content of the image becomes a multi-keyword description. Finally, on the basis of image retrieval, the texture and color of image visual content feature are extracted, and the color transfer method is improved by grasping the content characteristic of image. On the one hand, the texture similarity between images is calculated. In order to improve the success rate of coloring, color source images with similar contents to gray-scale images can be searched. On the other hand, color features can be extracted for more scientific pixel sampling, and the color transfer algorithm itself can be improved to enhance the effect of coloring.
【学位授予单位】：大连海事大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TP391.41

【参考文献】