Semantic Retrieval with Enhanced Matchmaking and Multi-Modality Ontology

This paper presents a method for automatically annotating and retrieving animal images. Our model is a multi-modality ontology extended from our previous works in the sense that both image feature and text information are taken into consideration. We use a medium size data set which contains animal images with their corresponding web pages. These images include 20 animal categories and are collected to form a systematic animal family. We are able to classify even close animal species which share similar appearances and infer their inner relationships from the family graph. This method also provides a ranking mechanism by calculating the ranking correlation in each modality to refine the the semantic matchmaking result. We assign ranking to the semantic relationships according the priority of their information. We show unequivocal evidence that our improved model achieves good accuracy. In this experiment we continue to concentrate on the domain of canine. These images are collected f...

• Type: Conference
• Year: 2007
• Where: ICMCS
• Authors: Huan Wang, Liang-Tien Chia, Song Liu
• Keywords: Animal, Animal Images, ICMCS 2007, We Believe