Abstract

Despite the widespread of test automation, automatic testing of graphical user interfaces (GUI) remains a challenge. This is partly due to the difficulty of reliably identifying GUI elements over different versions of a given software system. Machine vision techniques could be a potential way of addressing this issue by automatically identifying GUI elements with the help of machine learning. However, developing a GUI testing tool relying on automatic identification of graphical elements first requires to acquire large amount of labeled data. In this paper, we present Win GUI Crawler, a tool for automatically gathering such data from Microsoft Windows GUI applications. The tool is based on Microsoft Windows Application Driver and performs actions on the GUI using a depth-first traversal of the GUI element tree. For each action performed by the crawler, screenshots are taken and metadata is extracted for each of the different screens. Bounding boxes of GUI elements are then filtered in order to identify what GUI elements are actually visible on the screen. Win GUI Crawler is then evaluated on several popular Windows applications and the current limitations are discussed.