The Research and Implemention of Image Retrieval Based on User Interested Feature |
|
Author | LiuXiaoZhen |
Tutor | ZhangTianWen |
School | Harbin Institute of Technology |
Course | Computer Science and Technology |
Keywords | content-based image retrieval Web server log analysis user interested feature-based image retrieval |
CLC | TP391.41 |
Type | Master's thesis |
Year | 2008 |
Downloads | 120 |
Quotes | 0 |
The combination of Web and content-based image retrieval(CBIR) is gradually becoming the future study and development trend of content-based image retrieval system。In Web-oriented system, we get users interested feature by tracking the users’accessing records and retrieve with sample image. user interested feature-based image retrieval has effectively improved the efficiency of the users’search.In this paper, we completed the following work:(1) Accessing of user interested images. By web server logs analysis and information filtering, users delete useless web logs and web log domains , get the visit records and access to users interested image, and establish a visit index table of interest image , with users’intent.(2) Color and texture feature-based image retrieval. In this paper, we extract 20-dimensional color feature with HSV space based on the cumulative color histogram and 16-dimensional texture feature with Gray co-occurrence matrix, measure similarity between two pictures respectively with histogram intersection and continental distance,normalize the results and Synthesis the total similarity.(3) User interested feature-based image retrieval. It contains the following methods: Single-user interested feature based image retrieval and Multi-user interested feature based image retrieval.Single-user interested feature based image retrieval combine sample image feature and single-user interested feature, which is obtained by compositing user interested image and forgotten factor.Multi-user interested feature based image retrieval use sample and user interested feature, which is the result of multi-user interested feature clustering.In conclusion, the two methods could improve the efficiency of retrieval through experiments.