Trabalho em anais de evento

Leveraging Template Matching for Efficient Android Screenshot Visual Inspection

MYRRIA, ABDA; CASTRO, ALICE; MUNIZ, CAIQUE; BARBOSA, HERYCK; SILVA, ANDRÉ; CORDOVIL, LUIZ; PORTY, ROGER; SOUTO, THIAGO

2026 • 2026 IEEE International Conference on Consumer Electronics (ICCE) • 1 citações

Resumo

Graphical User Interface (GUI) object detection is crucial for applications such as GUI automation, testing, and code generation within the software industry. However, it presents unique challenges due to large in-class variance, high intra-class similarity, and the need for precise object localization. While deep learning has been a dominant approach, recent studies suggest that traditional pattern recognition algorithms or hybrid methods may outperform DNN models in certain scenarios. This study explores the application of Template Matching for automating the validation process of the Home Host Test Case within the Google Mobile Services (GMS) Test Suite (GTS). Our objective was to assess whether non-linear transformations are necessary to enhance template differentiation in this context. Surprisingly, our findings indicate that linear metrics provide sufficient discriminant capability, negating the need for additional complexity. This result highlights the efficacy of conventional template matching techniques in distinguishing visually similar icon templates within our test case framework, offering a robust baseline for detecting standardized GUI elements with high in-class variance. This study underscores the potential of traditional computer vision methods to streamline certification workflows and enhance efficiency in Android software quality assurance processes, reducing manual intervention.

Autoria