基于图像的多样化景点搜索

发布时间：2018-11-20 15:46

【摘要】：随着互联网带宽的迅猛提升，网络上的多媒体信息越来越多，其中景点相关的多媒体信息占了很大一部分比重，如旅游攻略文章以及旅游照片等等。鉴于广泛的应用前景，景点相关的多媒体信息越来越受到研究者们的关注，如何迅速地从这些海量的景点相关多媒体信息中获取和挖掘出满足用户需求的信息是信息检索和数据挖掘领域一直在努力解决的问题，景点搜索系统的提出旨在搜索和挖掘满足用户需求的景点信息并且通过很好的组织展现给用户。传统的景点搜索是基于文本的景点搜索，大多通过索引多媒体信息的周围文本信息比如文本描述以及标注来实现景点搜索。这种类型的景点搜索系统能很好地满足那些有明确旅游目的地的用户的需求，但并不是所有用户都对自己要去的旅游目的地有一个非常明确的描述，而且基于文本的景点搜索系统只允许用户将文本作为输入，这也给用户在使用时带来了很多的不便，，系统不能很好地理解用户的查询意图，这时就需要基于图像的景点搜索系统。比如，用户可以上传一张景点图片甚至画出他理想的旅游目的地，系统将会为用户搜索和输入图片相似的景点信息。虽然现在的部分搜索引擎已经提供了基于图像内容的图像搜索功能，但搜索结果仅仅是图片（以及包含图片的网页）而不是景点相关的多媒体信息。现有的景点搜索系统虽然能满足一部分用户的需求，但还是有很多需要加以改进的地方。基于文本的景点搜索只能允许用户输入关键词进行查询，基于文本的搜索容易存在关键词歧义等问题，不能很好地理解用户的输入并且用户也不能很好地表达查询意图；现有的景点搜索系统并没有对景点信息作很好的整理，用户不能对景点有一个直观的多方位的了解，景点摘要信息缺乏代表性，没有很好的覆盖度以及多样性；再者现有的景点搜索系统返回的景点搜索结果由于基于关键词匹配的原因往往都是没有整合的景点列表（或者相关网页），存在视觉冗余，更是缺乏多样性，没有很好的用户满意度。本文提出并构建了基于图像的多样化景点搜索系统。基于图像的多样化景点搜索系统允许用户上传图片作为输入，为用户返回多样化的景点搜索结果，并为景点给出代表性的景点摘要信息。系统的构建基于海量的景点图像集合，定性以及定量的实验充分验证了本系统的有效性，实验结果表明本系统很好地解决了传统景点搜索系统存在的不足。基于图像的多样化景点搜索系统的提出旨在解决现有景点搜索系统中存在的很多不足之处，能够让用户更好地表达查询意图、给出更加多样化的景点搜索结果以及更好的景点摘要信息。
[Abstract]:With the rapid improvement of Internet bandwidth, there are more and more multimedia information on the network, in which the scenic spot related multimedia information accounts for a large proportion, such as travel strategy articles and tourism photos, and so on. In view of the wide application prospect, the researchers pay more and more attention to the multimedia information related to scenic spots. How to quickly obtain and mine the information to meet the needs of users from these massive scenic spots related multimedia information is a problem that information retrieval and data mining field has been trying to solve. The purpose of the Scenic spot search system is to search and mine the scenic spot information to meet the needs of the users and to show it to the users through a good organization. Traditional scenic spot search is text-based scenic spot search, which is mostly realized by indexing the text information around multimedia information such as text description and tagging. This type of scenic spot search system can well meet the needs of those who have a specific destination, but not all users have a very clear description of the destination they are going to. Moreover, the text-based scenic spot search system only allows users to use text as input, which also brings a lot of inconvenience to users, and the system can not understand the user's query intention very well. An image-based scenic spot search system is needed. For example, if a user can upload an image of a scenic spot or even draw his ideal tourist destination, the system will search for information similar to the image for the user. Although some search engines now provide image search capabilities based on image content, the search results are only images (and pages containing images) rather than scenic spots related to multimedia information. Although the existing scenic spot search system can meet the needs of some users, there are still many areas that need to be improved. Text based scenic spot search can only allow users to input keywords to query, text based search is easy to exist keyword ambiguity, can not understand the user's input and users can not express the query intention well; The existing scenic spot search system does not have a good collation of the scenic spot information, users can not have an intuitive multi-directional understanding of the scenic spots, the scenic spot summary information is not representative, there is no good coverage and diversity; Furthermore, the search results returned by the existing scenic spot search system are often due to the lack of integration of the list of scenic spots (or related web pages) due to keyword matching, and there are visual redundancy and lack of diversity. No good user satisfaction. In this paper, we propose and construct an image-based multi-site search system. The system allows users to upload pictures as input, return the search results for the users, and provide the representative scenic spot summary information for the scenic spots. The construction of the system is based on a large number of scenic spot images, qualitative and quantitative experiments fully verify the effectiveness of the system. The experimental results show that the system can solve the shortcomings of the traditional scenic spot search system. The proposal of the image based diversified scenic spot search system aims to solve many shortcomings in the existing scenic spot search system and enable users to express their query intention better. More diversified search results and better site summary information are given.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.41

【相似文献】