Hybrid Approach of Using Visual and Tag Information for Situation-Oriented Clustering of Sightseeing Spot Images

Recent trend on the web is to share their traveling experience via uploading photos to web albums. Shared photos of sightseeing spots are important resources for those who are going to visit there. As sightseeing spot scenes vary with different situations, such as weather and season, automatic classification of photos into different situations is expected to be beneficial for tourists to plan when to visit there. This paper proposes a hybrid approach of combining content-based image clustering with filtering based on tag information of image. By using geotag information when retrieving images from web albums, collected images can be limited to a reasonable boundary to eliminate outliers. Content-based image clustering groups collected images into night, sunrise/sunset, cloudy, and shine situations. Moreover, by using the timestamp of images, the four situation categories are further verified to increase the accuracy. Experimental results show that the hybrid approach of content-based image clustering and tag-based filtering is effective for obtaining clusters with high precision and recall.