Reducing False Positives in Automated OpenCV-based Non-Native GUI Software Testing

Abstract

This paper is aimed at improving mobile game non-native GUI testing. We follow an approach to use OpenCV image recognition algorithms for detecting and accessing the hand-drawn GUI elements on the screen, in order to interact with them from within automated test scripts. In the previous work, we experienced the problem that some tests fail not due to the defects of the tested software itself, but because of the false positive results of template matching. It means that the high scores are sometimes elicited for the best match, though the requested GUI element is actually not present on the screen. In this contribution we investigate the possibilities of image filtering in order to reduce the number of such false positive cases. We describe our experiments with two algorithms supported by OpenCV library, a selection of GUI elements and mobile game scenes, and a number of image filtering methods. We demonstrate that using Canny edge detection filters can significantly improve the accuracy of recognizing false positive cases without affecting the true positive situations. Our conclusions can be helpful for improving hand-drawn GUI based mobile software testing reliability.