Font Size: a A A

The Research On Visual Semantic Concept Relationship

Posted on:2016-03-26Degree:MasterType:Thesis
Country:ChinaCandidate:W Y CaoFull Text:PDF
GTID:2308330473957047Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
With the rapid development of internet, image, audio and video have become the information carrier of people’s daily life. There are more and more research topics with the multimedia as the research object focused by industry and academia, such as image annotation, image classification, image retrieval, target tracking, target recognition, speech recognition and so on. Because of the characteristics of multimedia object is large and unstructured, which makes the multimedia object very difficult in the storage and retrieval. The most two famous of the many difficulties are the semantic gap and intention gap, which have also been the problems that researchers try to solve in the recent decades.The paper presents various types of concept correlation measurement method and by using the concept correlation measurement method as the concept relationship feature of multi category concept relation model to find various conceptual relations existing in the internet. This paper is a creative concept expression, also has the innovation of high-level semantic concept relation model. The published article also relates to the challenges of image expression, concept expression and concept correlation measurement, and provides an effective reference for the establishment of an automated scalable multi category visual semantic concept relationship network. For the concept relationship discovered in this paper can be used in the context based concept fusion method to improve the performance of image annotation, image classification or used in the user manual image annotation, user manual image retrieval to provide alternative keywords. Specifically, the results of this paper are listed as follows.1. This paper presents a variety of visual concept correlation measuring methods based on different concept expression, and the visual concept correlation measuring method based on visual words shows more effective in the concept similarity measurement. Semantic concept mainly includes a fixed number of visual words and a fixed number of representative images in the concept expression. At the same time, we found that adding with spatial relation of visual phrase in visual word is more effective than the individual visual words.2. In this paper concept relationship metric not only depends on the concept expression based on visual feature, but also extracts in the text representation and co-occurrence frequency of images. In the text representation, we measure the concept relationship by comparing the difference of the word element structure of text representation of concepts. In the co-occurrence frequency, we measure the concept relationship by computing the frequency of different concepts labeled in the same image.3. In this paper, we extract concept relationship from the visual image, the text representation of concept, image retrieval log this three aspects and establish the concept relationship model to predict the specific relationship between any two concepts. Through the identification concept relation of the model, we can recommend the related search terms to the user to reduce the intention gap existing between the intentions of the user and the retrieval words, and in the stage of image expression, we can combine the low-level image feature and the relationship between annotated concept of image to generate the new image feature to reduce the semantic gap existing between the low-level image feature and the high-level semantic concept.
Keywords/Search Tags:concept expression, concept correlation measurement, concept relationship feature, concept relationship, concept relationship network
PDF Full Text Request
Related items