Anime character face detection and recognition

• Main Author: Ng, Hui Chin
• Format: Final Year Project / Dissertation / Thesis
• Published: 2022
• Subjects: Q Science (General); T Technology (General)
• Online Access: http://eprints.utar.edu.my/4660/1/fyp_CS_2022_NHC.pdf; http://eprints.utar.edu.my/4660/
• Institution: Universiti Tunku Abdul Rahman

Intelligence (AI) applications have grabbed headlines with their potential to radically improve human life and even transform the industry. Currently, in the field of pursuing detection and recognition of objects, human face recognition is a welldeserved representative and has reached the highest level. Nevertheless, the use of AI in the field of art is still very few. The only famous ones nowadays are the appraisal of paintings and recognition of painting style. Let alone facial recognition in the field of art, such as the domain of this project, anime character face. There is still no mature technology or well-known application for recognising anime faces. Even the Google Lens, which claims to be able to search for all objects in the world, cannot identify an anime character with half accuracy. Thus, there is an acute need for such a system to detect and recognise anime character faces. Therefore, in this project, an AI mobile app named “Gease” with the main functionality of detecting and recognising multiple anime character’s faces are developed. The mobile app will be the first mobile app specifically for anime character face detection and recognition. Likewise, the recognisable number of characters is up to 205 which are more than doubled compared to most of the existing anime face detection and recognition projects which only involve at most 100 characters. The accuracy achieved 97% and 63.2% Top-1 accuracy for detection and recognition respectively. While the inference time is 31.9ms and 4.8ms per face detection and recognition respectively. Moreover, the FastAPI web framework that supports Python development was utilized in this project. FastAPI allows the app developed to be opened to external connections as a REST API and return the detection and recognition results in JSON form which greatly flexes and expands the value of the app. Furthermore, this project developed an exquisite and smooth user interface while having multi-platform adaptability. The mobile is a portable PWA which flourishing recently. It accommodated multiple platforms including web browser, Android and iOS with only one set of codes but maintains most of the functions of native apps. The Ionic framework was used in design a cross-platform mobile user interface.