Image Clustering Based on Correlation Between Visual Features and Annotations
Tianwen Zhang · Dianzi xuebao · 2006
The paper proposes an unsupervised semantic categorization algorithm for annotated images.In order to establish image categories automatically by unsupervised learning,the algorithm first scores the annotation words for each image by using their relevance to visual features.The scores of annotation words indicate to what extent these words have visual characteristics.The words with a good visually discriminative power can be chosen as image categories.Then a recursive clustering algorithm is presented to group images into the extracted semantic categories according to their annotation.Experiments using a 4500-image Corel database show the efficacy of the proposed algorithm.The results can be exploited for better image browsing and image retrieval.