Relevance feedback is important in bridging the semantic gap between the low-level visual features and the high-level semantic concepts during the image retrieval. In this work, the image retrieval using relevance feedback involves four phases. In the first phase, an initial retrieval is carried out for a given query image and each of the retrieved images is rated as relevant or irrelevant. In the second phase, the rated images are clustered. In the third phase, a score is computed for each image in the repository to measure the degree of its relevance to the query using the images in the clusters. We propose a method to compute the score using the local relevance feedback and the global relevance feedback. The local relevance feedback based component of the score is computed using the instance-based feedback approach. In this approach, the degree of relevance is measured using the feedback from the current iteration only. The global relevance feedback based component of the score is computed using the query-point movement approach. In this approach, the degree of relevance is measured using the feedback from different iterations. The scores for the images in the repository are used to retrieve the images. Each of the newly retrieved images is rated as relevant or irrelevant. In the fourth phase, each of the newly rated images is assigned to an existing cluster of rated images. We propose to represent an irregularly shaped cluster using multiple representatives. These representatives are used to assign each of the newly rated images to a cluster. The third and fourth phases are repeated until there is a convergence of the process of retrieval. The image retrieval performance for the proposed methods is compared with that of the existing methods on Wang and Corel image datasets. Results of these studies demonstrate the effectiveness of the proposed methods in improving the retrieval performance. Copyright 2014 ACM.