An image retrieval approach with relevance feedback

An image retrieval approach combined with relevance feedback is proposed. A set of blobs that are generated from image features using clustering can be used to describe an image. Given a training set of images with annotations, we apply probabilistic models to predict the probability of a blob in image according to the query words. For improving the initial query results, we apply a relevance feedback mechanism to bridge the semantic gap, leading to the improved image retrieval accuracy. A support vector machine classifier can be learned from training data of relevance images and irrelevance images labeled by users. Experimental results show that the proposed approach obtains higher retrieval accuracy than a commonly used approach.

I. Introduction

The number of digital images and videos has grown exponentially because of the popularity of economical imaging devices and the increasing capacity of databases. Large image collections are now available and shared online. Therefore, it is becoming more difficult to manage and retrieve images, how to retrieve image is an active area of research in the fields of information retrieval and computer vision in recent years.