Image Retrieval Using Search Theory

With the rapid development of modern Internet, many kinds of data growexplosively. The efficient data retrieval methods are exigent. Content-Based imageretrieval is an effective way which avoids subjective effects and human resource cost inkey word based methods. Research on image retrieval now is focused on how to extractfeatures effectively since image retrieval will get better results if the feature candescribe more information about the image. In real image retrieval system, the speed ofimage retrieval can not meet the demand with huge amount of data. It can not search allimages in a limited time. This thesis focused on resource allocation plan which appliedsearch theory so that the image retrieval can meet the demand in a limited time. SearchTheory is created by B. Koopman in1940s, it has been applied to many fields since theSecond World War, such as oil exploration, market research and search lost people orthings. In recent years, it has been applied to bioinformatics, intrusion detection andwireless sensor networks.In this thesis, the main research work is as follows:1. The effect of features based on color or shape on image retrieval is discussed,such as color histogram, edge histogram, invariant moments, and contour descriptor.Multiple features are used in a retrieval and SVM is used for feedback.A prototype ofimage retrieval system has been implemented.2. Optimal search theory is applied to establish a model which can select a subsetof the total data in a limited time to retrieval image. It makes the maximum possibilityof getting the image satisfied the demand of similarity and do not beyond the predefinedtime. The image classification and initial possibility estimate are discussed. The regulardetection function is found and the error analysis of exponential type function is made.An unregular detection function is found. Its solution and parameter determination arediscussed. The optimal plan and non-optimal plan are compared through experiment.3. A more genral search model is proposed which can be applied to distributedsituation and a certain case is examined to which the optimal search theory can apply.