Computational Cartographic Recognition

Résumé du livre

Maps play an important role in providing geographic-related information and explanations regarding topics of interest. Maps are artifacts that are made by humans and, more importantly in this context for humans to read. While humans can develop map reading skills to comprehend qualitative and quantitative information from maps, can computers recognize information from maps and understand them as we do? In this dissertation, we broadly refer to the ability of computers to recognize the information on map images as computational cartographic recognition. Recent advances in the field of computer vision have shown that artificial intelligence and machine learning methods can be used to successfully recognize and classify a wide range of images. The dissertation research represents preliminary steps toward computational cartographic recognition, aiming to explore how these methods can be used to recognize information from maps. There are three research objectives achieved in the dissertation. First, we use machine learning methods to recognize fundamental cartographic information of maps including the geographic region mapped and projection used on the map. The limits of the methods are also examined when maps are presented with different degrees of distortions. Second, we develop deep learning-based models to recognize themes from map titles or legend titles of choropleth maps and classify the themes based on their semantic meanings. Themes are important for map users to understand the contents on maps because a theme indicates what phenomenon is presented on a map. Third, to explore whether computers can recognize spatial patterns as humans do, we develop a computational framework to recognize spatial patterns on choropleth maps. We also conduct a survey on how humans read spatial patterns on choropleth maps and compare the survey results with those from the computational models. The results for the three research objectives suggest that the models developed for the tasks are capable of recognizing information from maps, but there are also limitations of the models.