Study on the content-based image retrieval system by unsupervised learning

The content-based image retrieval (CBIR) system aims at searching and browsing the large image digital libraries based on automatically derived imagery features. This paper introduces two algorithms based on the normalized cut for images clustering. We extract the color and texture features for computing the distance between the images, and take advantage of the bipartition method and minimum spanning tree for grouping. The performance of this system using the above methods is evaluated on a database of around 8000 images from the internet. The searching accuracy is satisfied for the target requirement.