Image clustering based on semantic similarity

Abstract

Image clustering is an interesting field in machine learning and computer vision, in which images are classified into a set of similar groups. Recently, with the explosive growth of the data in the smartphone and the web (Facebook, Instagram…), image clustering has even been a critical field to help the user quickly access the visual information he is looking for. Existing methods of image clustering only used either low-level visual feature, which constitutes a major obstacle to obtaining an accurate set of similar groups. To tackle this problem, we propose a novel algorithm that can cluster images based on the semantic similarity between surrounding texts (concept) of each image. In particular, we group images depending on the semantic similarity of their concepts instead of visual similarity. Conclusively, images are automatically clustered based on the label features. The performance of the cluster was compared based on accuracy. The highest accuracy was obtained by applying the method of Lin with 88.89%.