An enhanced relevance feedback method for image retrieval
The rapid growth of the computer technologies and the advent of World-Wide Web have increased the amount and the complexity of multimedia information. Images are the most widely used media type other than text to retrieve hidden information within data and it is used as a base for representing and r...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2008
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/9535/1/LimPeiGeokMFSKSM2008.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The rapid growth of the computer technologies and the advent of World-Wide Web have increased the amount and the complexity of multimedia information. Images are the most widely used media type other than text to retrieve hidden information within data and it is used as a base for representing and retrieving videos, flash and other multimedia information. An efficient image retrieval tool needs to be developed to select the appropriate images from a digital images database in response to user queries. A content based image retrieval (CBIR) system has been proposed as an efficient image retrieval tool which the user can provide their query to the system to allow it to retrieve the user’s desired image from the image database. However, there are several problems have been identified by previous researches such as semantic gap between high level query to low level features and human subjectivity. Therefore, relevance feedback mechanism has been introduced to integrate with CBIR system which intends to solve the problem of CBIR and indirectly increase the CBIR performance. Unfortunately, the traditional relevance feedbacks have some limitations that will decrease the performance of CBIR. In this study, the imbalance training set issue has been highlighted. Imbalance training set is an issue that the negative samples are overwhelming the positive samples during the relevance feedback process. As a result, insufficient training occurs and further degrades the performance of CBIR. To solve the problem, a representative image selection and user weight ranking methods have been introduced. Besides that, Support Vector Machine (SVM) has been proposed as a technique to aid the CBIR learning process. Through the learning process, the system will be able to adapt to different circumstances and situations. Finally, the experiment results reveal that the proposed method is better than traditional relevance feedback method which success improves the performance of CBIR. |
|---|