Relevance feedback and region based image retrieval are two effective ways to improve accuracy in content-based image retrieval. In this paper, we propose a content-based image retrieval method using relevance feedback and homogeneous region. By extracting a number of homogeneous color regions from the image and calculating the occurrence frequency of regions, we convert image feature vectors to weighted vectors. On the basis of the weighted vectors, we calculate the similarity between two weighted vectors and using relevant feedback technique. Our experimental results on a Wang database of over 10,000 images suggest that the technique results in which is close to user's intention better than the CBsIR and CCH methods.  Index Terms-content based image retrieval, weighted vectors, feature vectors, machine learning.

There are two commonly used aggregation based approaches in Content Based Image Retrieval (CBIR) systems using multiple features (e.g., color, shape, texture). In the first approach, the systems usually represent each image as a unified feature vector by concatenating component feature vectors and then for a query image, compute its distance measure with images in the database. Inspite of the simplicity, this approach does not emphasize the importance of each component feature. Another approach often computes the weighted linear combination of individual distance measures and the weight assignment to each is based on Relevance Feedback (RF) from a user to determine the importance of each component. In this paper, the authors propose to use Pareto approach for candidate selection. The proposed algorithm produces a compact set of candidate images when comparing with the entire dataset and this set also contains results obtained from all aggregation operator [3]. The authors also formalize main properties of Pareto front with respect to CBIR which are mainly used to propose our two algorithms. The experiments on three image collections show that our proposed approach is very effective to improve the performance of the classification engines.