An Ensemble Approach for Visual Testing of Web Applications

A novel ensemble approach for visual testing of web applications is proposed here, which aims to ensure the visual consistency and functionality of web applications across diverse configurations and user demands. The proposed approach integrates three models: a contour-based detection model, a YOLO-based granular analysis model, and a CNN-based design comparison model. The contour-based detection model identifies pixel-wise differences between the input and reference screenshots, the YOLO-based granular analysis model segments the screenshots into UI elements and locates the discrepancies, and the CNN-based design comparison model evaluates the aesthetic similarity between the input and reference screenshots. The proposed approach uses a comprehensive dataset of web application screenshots from Roboflow, comprising various UI elements and annotations. The paper presents empirical studies that show the superior performance and effectiveness of the proposed approach in detecting visual discrepancies and preserving user experience. Exploring the challenges and limitations of existing visual testing methods and tools, the study also outlines future directions to enhance the quality and reliability of web applications. Contributions to the advancement of automated testing techniques and tools for web development are made through this work.