The Application Of Reinforcement Learning And Relevance Feedback In Orthodontics Image Retrieval

A large number of images come forth because of the rapid development of the multimedia and network techniques. Traditional text-based image retrieval approach can't adapt to the demand of image data retrieval, so content-based image retrieval (CBIR) technique becomes the current research focus. To get a more accurate result, relevance feedback technique is introduced into image retrieval and proved to get a nice human-machine interactive. Orthodontics image retrieval is an important part of orthodontics. Effective retrieval has made for the fast and accurate orthodontics treatment and also improved the rate of success.The core of CBIR technique is to express the features of the images. This paper describes the extraction of three main image features: color feature, texture feature and shape feature. Then it describes the matching technologies. While describes extraction of the image features, introduces the three color spaces of color image and their conversion formula in details, and focuses on several main methods of each feature extraction. After that, gives the introduction of the image matching technology as well as the evaluation criteria of image retrieval system. In the later experiments, according to the characteristics of orthodontic images, this paper selects the retrieval method technique combined color feature and shape feature. While carries on the color feature extraction, it uses the color histogram based on the HSV space, and also the shape feature extraction uses the Hu invariant moment features. Experiments shows that the image retrieval technique combined color and shape features overcomes the limitations of the single feature retrieval technique, and makes the image retrieval more comprehensively and effectivelyThe introduction of relevance feedback to the image retrieval system improved the efficiency. Based on the detailed introduction of the theories about relevance feedback and reinforcement learning, this paper introduces the reinforcement learning to the relevance feedback, proposes a relevance feedback method combined the Q-learning and Bayesian theory. With the orthodontics image base, experiment is carried out by matlab7.0.1 and shows the algorithm's efficiency. Lastly,the paper introduces an simple image retrieval system based on relevance feedback which realizes using the Matlab'GUI .