English title

Content-based image retrieval (CBIR) has received considerable research interest in the recent years. The basic
problem in CBIR is the semantic gap between the high-level image semantics and the low-level image features.
Region-based image retrieval and learning from user interaction through relevance feedback are two main
approaches to solving this problem. Recently, the research in integrating these two major techniques has gained
many attentions .The representative is the relevance feedback based Multiple Instance Learning mechanism. This
paper presents an interactive content-based image retrieval system that incorporates Multiple Instance Learning
(MIL) into the user relevance feedback to learn the user’s subjective visual concepts. The proposed model consists
of three main components: The transforming unit (bag generator), the learner unit, and the retrieval unit. In the
transforming unit, each image of the database is transformed into the corresponding image bag. The learner unit uses
these bags, user's relevance feedbacks and the proposed MIL method to learning user-interested visual concepts. In
the retrieval unit the images of the database are ranked using a two-phase ranking algorithm. Our model is designed
for using in applications that need image retrieval based on the general structures of images such as scene
classification and retrieval. We have tested our model on a natural scene image database, consisting of 3000 images
taken from the COREL library. Performance is evaluated and the effectiveness of the proposed model has been
shown through comparative studies.