Location:Home > Engineering science > Computer Science > Computer System Architecture > Research and Implementation of the Image Search System Based on Interest Region Matching

Research and Implementation of the Image Search System Based on Interest Region Matching

Downloads: []
Tutor: LiuHuiLin
School: Northeastern University
Course: Computer System Architecture
Keywords: DM-Watershed algorithm,interest region,similarity measurement,image search
CLC: TP391.41
Type: Master's thesis
Year:  2009
Facebook Google+ Email Gmail Evernote LinkedIn Twitter Addthis

not access Image Error Other errors

As the computer technology involves, content-based image search system have been a good approach to search images, to some extent. However, because the content-based search system rely on global features such as color, texture and shape to describe an image, whereas those features may not necessarily represent the users¡¯regions of interest, the search result is rather limited. Thus, the image search system based on interest region matching greatly improves the accuracy and target-coverage of image search, for it aims at the specific regions that users are interested in rather than global features. This method is gaining more and more attention by the day.The essay analyzed the workflow and difficulties in image search, designed a search system structure based on interest region matching and put forward a dynamic region combined DM-Watershed division algorithm, through researches on the classic division algorithms. The DM-Watershed algorithm is composed of region division and dynamic region combination, the first of which divide the image by adding a target region mark to the Watershed algorithm, and the later solved the problem of over-division by utilizing the idea of dense-region-oriented expansion in OPTICS cluster. The essay also extracted from grayscale mutation interest regions using Wavelet transform and bring up with methods for interest region similarity measurement. Image division algorithm and similarity measurement were applied to the image search system, implementing several function models of system pretreatment, image region division, interest region recognition, interest region feature extraction and similarity measurement.Lab result shows that this system is capable of dividing region accurately and searching the correlative images almost fully, greatly enhancing the efficiency and target-coverage in comparison with global-feature based image search systems. The capability as a whole is able to meet the demands of most applications and is of great practical value.
Related Dissertations
Last updated
Sponsored Links
Home |About Us| Contact Us| Feedback| Privacy | copyright | Back to top