Dissertation > Excellent graduate degree dissertation topics show

Research on Web Image Retrieval Based on the Fusion of Textual Information and Visual Information

Author: ZhengXin
Tutor: ChenLian
School: Nanchang University
Course: Computer technology
Keywords: Web image retrieval semantic similarity image auto-annotation text information visual information Bayesian inference networks and information integration
CLC: TP391.41
Type: Master's thesis
Year: 2009
Downloads: 56
Quote: 0
Read: Download Dissertation

Abstract

With the computer technology and the rapid development of Internet, various information is stored on the Internet, this huge library of information. As the Internet, resource sharing and the power of fast communication, making the increasing Internet penetration, there is increasing use of Internet search and search all kinds of information, of which not only have simple text data, and more, including a large number of images, video and other multimedia information. The image as a multimedia information on one of the important present form, it is through color, texture, shape and other characteristics of a rich visual, intuitive, and lively manner so that abstract data visualization, real-oriented to be presented to the general public, and promoted between people exchange of information will help in-depth knowledge and understanding. However, the image semantics rich, semantic understanding is different from the text, the lack of clear criteria for writing semantic understanding by people’s perception and knowledge of structure factors, different people the same semantic understanding of the image there are deviations, how fast, accurate Web resources from the massive search for a user interested in the image into a very challenging task. The existing major Web image retrieval method can be divided into two kinds: one is text-based image retrieval (TBIR), the other is content-based image retrieval (CBIR).Text-based image retrieval method has a huge workload of manual annotation, and text annotation strong subjectivity can not fully cover the content of the image itself is flawed and that content-based image retrieval solution may be effective to overcome the subjectivity of manual describe the image to improve image retrieval efficiency, but the habit of human beings for image recognition image-based high-level semantics, the use of images to express the underlying statistical characteristics of high-level semantics of images there are some limitations, therefore, content-based image retrieval, only the bottom of image-based features of search, increasing the user the complexity of the retrieval operation,Ordinary users do not meet the general cognitive habit for image.To solve the above two kinds of image retrieval methods of its own defects and the characteristics of Web images, this paper presents an improved Web image retrieval methods: the images from the Web page where the text information contained in the text to extract features, and low-level visual images from the Web Feature extraction of high-level semantic features in a combination of Web image retrieval.Based on the above proposed image retrieval method, first of all, this paper semantic similarity computing technology as a measure of semantic information means to measure Web image similarity of the Chinese version of the information, and research behind Web image text information and visual information (that is, from low-level visual features to extract high-level semantic features) in combination provide the basis and platform.Second, Web images of low-level visual features and high-level semantic features insurmountable exists between the "semantic gap" problem for the semantic gap, this paper presents a Web-based image content classification auto-tagging method to extract the image of high-level semantic feature ; Then, using semantic similarity computing technology, to measure the extracted semantic feature of high-level quality, and further high-level semantic image content features and Web image text messaging better together.Then, in order to take full advantage of Web images extracted from the text information and images from the Web to extract low-level visual features of high-level semantic features, some of the content of these two sources of information on the multi-fusion ability of Bayesian Inference online, their be fully integrated together to achieve the text information and visual information based on a combination of Web image retrieval method.Based on the above, this paper designs and implements a Web image retrieval prototype system to extract the image content from the Web high-level semantic features, and text information from a Web image to extract the text features into the system, so that the two combined to realize the full Web image retrieval, the results validate the method proposed in this paper in the Web image retrieval more effective.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile