Detecting In-Game Toxicity via Bullet Hole Patterns Using Image Recognition
Rantanen, Onni-Petteri (2024)
2024
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2024091825310
https://urn.fi/URN:NBN:fi:amk-2024091825310
Tiivistelmä
The purpose of this thesis was to develop a system capable of detecting toxic messages
conveyed through bullet hole patterns in first-person shooter games, with a focus on
Counter-Strike 2. The thesis sought to address the challenge of identifying inappropriate text created by players within the game environment, which goes undetected by traditional moderation systems. The research questions centered around determining the most effective image preprocessing techniques and Optical Character Recognition (OCR) parameters to accurately detect and recognize text formed by bullet holes. The broader issue of in game toxicity, including various subtle and overt behaviors, was also explored. This thesis was commissioned as a personal project motivated by the author’s experiences with in-game toxicity.
The research problem was approached through a practical, experimental methodology. The central concepts related to image recognition and OCR were explained first. The thesis then discusses the development and testing of various image preprocessing techniques, including Gaussian Blurring, Bilateral Filtering, and Contour Detection. The primary research method involved iterative testing of these techniques to evaluate their impact on the accuracy of the Tesseract OCR engine in recognizing text under different conditions. Data was analyzed by comparing the effectiveness of these methods in various game scenarios to identify the most reliable approaches. Research demonstrates that no single preprocessing technique is universally effective for all scenarios. Instead, the success of OCR depends on the specific case, requiring adaptive preprocessing and carefully selected OCR parameters. The analysis indicates that automating the selection of preprocessing techniques and OCR settings could significantly enhance the system’s use. This project lays the groundwork for further exploration into more sophisticated and adaptable detection systems for combating in game toxicity.
conveyed through bullet hole patterns in first-person shooter games, with a focus on
Counter-Strike 2. The thesis sought to address the challenge of identifying inappropriate text created by players within the game environment, which goes undetected by traditional moderation systems. The research questions centered around determining the most effective image preprocessing techniques and Optical Character Recognition (OCR) parameters to accurately detect and recognize text formed by bullet holes. The broader issue of in game toxicity, including various subtle and overt behaviors, was also explored. This thesis was commissioned as a personal project motivated by the author’s experiences with in-game toxicity.
The research problem was approached through a practical, experimental methodology. The central concepts related to image recognition and OCR were explained first. The thesis then discusses the development and testing of various image preprocessing techniques, including Gaussian Blurring, Bilateral Filtering, and Contour Detection. The primary research method involved iterative testing of these techniques to evaluate their impact on the accuracy of the Tesseract OCR engine in recognizing text under different conditions. Data was analyzed by comparing the effectiveness of these methods in various game scenarios to identify the most reliable approaches. Research demonstrates that no single preprocessing technique is universally effective for all scenarios. Instead, the success of OCR depends on the specific case, requiring adaptive preprocessing and carefully selected OCR parameters. The analysis indicates that automating the selection of preprocessing techniques and OCR settings could significantly enhance the system’s use. This project lays the groundwork for further exploration into more sophisticated and adaptable detection systems for combating in game toxicity.