In this paper, we discuss the semiautomatic annotation that can build the semantic index for image retrieval systems. Every image will be annotated at annotators' private and partial wording, and then we use WordNet to add extra relative keywords to increase the searched probability of each image. Unfortunately, blind expansion will make the index of images unreasonable and unwise due to the common polysemous-characteristic of English words. Therefore, we use a modified Word Sense Disambiguation technique for the image annotation to identify the sense of every annotative keyword. After examining the results, we concluded that with the proposed mechanism, the precise rate of the image retrieval system has a dramatic improvement.
Proceedings of Taipei and the International Computer Symposium 2004，6頁