Background: Testing is an important part of developing a software project, and when the project is under continuous development, regression testing is an important part of that testing. Regression testing the logic of the software project is not that different from normal testing of the logic. Making sure that there are no unintended changes or visual faults in the UI is more complicated, and that is where visual regression testing can be used.

Objectives: The objective of this thesis is to document different approaches to visual regression testing. This includes documenting the most supported approaches to visual regression testing and their strengths and weaknesses. As well as testing how well these approaches would hold up to real-world use.

Methods: To accomplish this, two different literature reviews and an experiment were conducted. The first literature review is to determine which approaches are the most supported both in academia and in the industry. The second literary review is to determine the strengths and weaknesses of the most supported approaches. The experiment then consists of testing tools representing each approach on websites in high contrast mode to determine how many visual faults they can detect compared to manual detection.

Results. The two literature reviews resulted in the identification of three different approaches: Pixel to pixel comparison, AI comparison, and DOM comparison, with strengths and weaknesses for each, such as “Pixel to pixel comparisons” and DOM comparisons weakness in returning false positives. While one of AI comparison strengths was its reduced number of false positives. The result from the experiment was that none of the tools performed as well as manual detection, with both Pixel to pixel comparison and AI comparison returning a lot of false positives, while DOMcomparison did not identify a single fault.

Conclusions. The conclusion for this thesis is that the different approaches are best used in different projects based on the environment, and that DOM comparison is not suitable for visual regression testing on websites in high contrast mode.