Label Detection and Discrepancy Analysis across Map Providers using Computer Vision

Abstract

In this thesis we present a system that compares the text recognized from tiles images downloaded through the APIs provided by the different map providers, specifically in this thesis we compare Bing maps, Google maps, and OpenStreetMap. Then the system uses Computer Vision to understand the text from the tiles and help in reading it accurately through OCR.The system highlights the discrepancies in the text, where we review the words that appear in a specific area and then determine whether they appear or not in each map provider, and the system determines if they appear completely mismatched due to some misspelling or any other reasons. It also compares the word size system between each map provider, the color of the word, and its different location in each map provider. The system is distinguished by displaying a large number of statistics about missing words in each map provider by country, as well as about mismatched words, and displaying words that need revision to make their size larger or smaller compared to other map providers. This makes the tool valuable to the various geographic editorial teams for detecting, reviewing, and modifying text inconsistencies on the map providers.