Dissertation
Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

Application of Q-Learning in the Content-Based Image Retrieval Technology

Author GuoJin
Tutor GuoMaoZu
School Harbin Institute of Technology
Course Computer Science and Technology
Keywords Content-based Image Retrieval(CBIR) Feature Extraction Relevance Feedback Q-learning
CLC TP391.41
Type Master's thesis
Year 2008
Downloads 53
Quotes 0
Download Dissertation

Investigation of how to improve the search results to content-based image retrieval is always an emphasis and ultimate goal. There have been lots of algorithms in order to achieve this goal, such as feature extraction algorithms, similar measurement algorithm and relevance feedback algorithm, which have gotten some fruits.However, there has been not a standard set of image testing and a model of searching results evaluating from the existing searching algorithm, which can show a certain method is capable to be applied universally or get preferable searching result. Therefore it will effectively improve the retrieval capabilities if there is a dynamic selection method which can provide different image retrieval methods for different image databases. This paper proposes a method to solve this problem on the basis of further studying this idea, which introduces Q-learning algorithms into the image retrieval technology to choose retrieval methods for enhancing retrieval results.Specific, The main research and innovation are as follows:Firstly, this paper analyzes the three basic and commonly used relevance feedback algorithms including the main ideas, mathematical models, respective advantages, disadvantages, and scope of application, at the same time, the algorithms are programmed. On this basis of above, the paper introduces the thinking of integrating relevance feedback. To achieve the best retrieval results, the experiments discuss about how to set the parameters of relevance feedback.Secondly, this paper introduces Q-learning algorithm into the content-based image retrieval system to choose retrieval methods, at the same time, the prototype system and the achievement process of choosing feature extraction algorithm and relevant feedback algorithm are elaborated. The application of the idea about explanation-based learning is used to solve the issues of slow convergence for Q-learning algorithm, which can further improve the convergence rate.Thirdly, a Q-learning CBIR system platform is introduced. The Q-learning algorithm proposed in this paper is used in the system platform to choose retrieval algorithm for the specific image database.Finally, the content-based image retrieval system is on the prospect, and the further research direction of the field is discussed.

Related Dissertations
More Dissertations