부산대학교 도서관

Visual testing is a software testing technique that checks the visual aspects of a Graphical User Interface (GUI). It helps identify visual defects that other GUI testing techniques, such as functional and performance tests, may not detect. While functional testing verifies the correct behavior of applications, it is not effective in catching visual issues. This paper introduces an AI-driven visual tester, a novel approach to visual testing that enhances the detection of visual defects in application interfaces. The tester can effectively test any platform, using screenshots as input. The process begins with object detection in the screenshots, employing two methods: traditional image processing and deep learning. The detected objects are then transformed into a tree data structure for efficient analysis. Object matching is utilized to correlate objects between the reference and updated images. Finally, change analysis and classification are performed to identify and categorize changes in each object, such as translation, scaling, color change, or object removal.