ALBERT

Description: While visual appearances play a main role in recognizing the concepts captured in images, additional information can provide complementary information for fine-grained image recognition, where concepts with similar visual appearances such as species of birds need to be distinguished. Especially for recognizing geospatial concepts, which are observed only at specific places, geographical locations of the images can improve the recognition accuracy. However, such geo-aware fine-grained image recognition requires prior information about the visual and geospatial features of each concept or the training data composed of high-quality images for each concept associated with correct geographical locations. By using a large number of images photographed in various places and described with textual tags which can be collected from image sharing services such as Flickr, this paper proposes a method for constructing a geospatial concept graph which contains the necessary prior information for realizing the geo-aware fine-grained image recognition, such as a set of visually recognizable fine-grained geospatial concepts, their visual and geospatial features, and the coarse-grained representative visual concepts whose visual features can be transferred to several fine-grained geospatial concepts. Leveraging the information from the images captured by many people can automatically extract diverse types of geospatial concepts with proper features for realizing efficient and effective geo-aware fine-grained image recognition.

Description: Geographic information, such as place names with their latitude and longitude (lat/long), is useful to understand what belongs where. Traditionally, Gazetteers, which are constructed manually by experts, are used as dictionaries containing such geographic information. Recently, since people often post about their current experiences in a short text format to microblogs, their geotagged (tagged with lat/long information) posts are aggregated to automatically construct geographic dictionaries containing more diverse types of information, such as local products and events. Generally, the geotagged posts are collected within a certain time interval. Then, the spatial locality of every word used in the collected geotagged posts is examined to obtain the local words, representing places, events, etc., which are observed at specific locations by the users. However, focusing on a specific time interval limits the diversity and accuracy of the extracted local words. Further, bot accounts in microblogs can largely affect the spatial locality of the words used in their posts. In order to handle such problems, we propose an online method for continuously update the geographic dictionary by adaptively determining suitable time intervals for examining the spatial locality of each word. The proposed method further filters out the geotagged posts from bot accounts based on the content similarity among their posts to improve the quality of extracted local words. The constructed geographic dictionary is compared with different geographic dictionaries constructed by experts, crowdsourcing, and automatically by focusing on a specific time interval to evaluate its quality.