Contextual Image Classification towards Metadata Annotation of Thai-tourist Attractions

Author
Pakpoom Mookdarsanit, Lawankorn Mookdarsanit
Abstract
Tourism in Thailand is an economical travel, remaining the most popular program reviewed by the international tourist agencies. Due to an inexpensive cost of living in Thailand, many tourists choose this country as the holiday destination. Tourists could enjoyably spend their money on the various tourist attractions such as historical places, beach & nature places, cultural street & floating markets, health & medical tourisms, and modern places & shopping malls. Many tourists also share the identity of Thai in the form of “images” on the social media, making an online promotion of Tourism in Thailand. However, many images which have not provided any metadata of tourist attraction are inconvenient to find the information of unknown places, making the tousrists ignore Thailand, negatively affecting the national income. Using a fully supervised learning for Thai-tourist annotation, it also consumes a massive amount of training data. In this paper, an eager learning framework of contextual image classification is proposed to annotate the metadata of tourist attractions. From the comparative results, it indicates that a eager learning could perform better than the fully supervised learning.
Keywords
Tourist Attraction Recogntion; Image Annotation; Image Retrieval