Effective Image Annotation for Search Using Multi-level Semantics

Abstract

There is an increasing need of development of automatic tools to annotate images for effective image searching in digital libraries. In this paper, we present a novel probabilistic model for image annotation based on content-based image retrieval techniques and statistical analysis. One key obstacle in applying statistical methods to annotating images is the amount of manually-labeled images, which are used to train the methods, is normally insufficient. Numerous keywords cannot be correctly assigned to appropriate images due to lacking or missing in the labeled image database. We further propose an enhanced model to deal with the challenging problem. With the model, the annotated keywords of a new image are determined in terms of their similarity at different semantic levels including image level, keyword level and concept level. To avoid some relevant keywords missing, the model prefers labeling the keywords with the same concepts to the new image. Obtained experimental results have shown that the proposed models are effective for helping users annotate images in different training data qualities.