METHOD FOR REGION BASED ON IMAGE RETRIEVAL USING MULTI-CLASS SUPPORT VECTOR MACHINE

摘要

A method for region based image retrieval using multi-class SVM(Support Vector Machine) is provided to apply weight by the user's feedback on appropriateness decision, thereby influencing a user's intention. An proper image collection is generated by a user's appropriateness decision after outputting similar image by matching the user's request image data with data in the database(S26). A multi class SVM classification device is learned by generating the optimum region class at the first stage(S53). The obtained appropriate image repeating above process is classified using the multi-class SVM classification device, and refined region class is generated by combining classes with high similarity(S47). The multi-class SVM classification is learned with the refined region class, and the new requesting point is inputted.