Automatic Optical Character Recognition Preprocessing for Custom Gameplay Text

Computer Games are one of the key use cases of graphics cards of AMD. To ensure highest quality and performance, extensive testing of graphics hardware and software is required. However, much of this gameplay testing is manual and requires significant efforts due to varying styles in games and their versions. In this context, an open challenge lies in the difficult to automatically pre-process multiple heavily styled and color instances of text that appear in various games which current requires manual tuning. The goal of this project is to investigate and implement machine learning-based solutions to automatically pre-process, detect and recognize text in gameplay settings. The solution developed should be able to handle varying art, text, lighting, and user-interface styles and configurations as seen in games. The development of this framework would result in saving a significant amount of manual effort needed for automated testing of games.

Faculty Supervisor:

Babak Taati

Student:

Partner:

AMD Canada

Discipline:

Computer science

Sector:

Manufacturing; Professional, scientific and technical services

University:

University of Toronto

Program: