Article Per Year (5 Year)
p-Index From 2020 - 2025
0.23
P-Index
This Author published in this journals
All Journal JOIV : International Journal on Informatics Visualization
Godata, Gempar Bambang
Unknown Affiliation
Computer Science & IT

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

 1 
Web-Based Deep Learning Approach to Identifying AI-Generated Anime Illustration Johan, Monika Evelin; Wong, Richard Faustine; Godata, Gempar Bambang; Wijaya, Westley; Haezer, Eben
JOIV : International Journal on Informatics Visualization Vol 9, No 4 (2025)
Publisher : Society of Visual Informatics

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.62527/joiv.9.4.2899

Abstract

As technology advances rapidly in artificial intelligence, the dominance of generative artificial intelligence (AI) images becomes increasingly evident in art, design, and the creative industry. However, the generative AI has processed numerous images from the Internet, including copyrighted content, trademarks, and artists' illustrations, which pose legal risks. Consequently, the manual tasks involved in managing and classifying these images have become more complex and time-consuming. Therefore, this research proposes the application of deep learning techniques, specifically Convolutional Neural Network (CNN), to automate the process of classifying AI-generated illustrations. The research was conducted by the Cross-Industry Standard Process for Data Mining (CRISP-DM) method. Initially, the study began with a literature review to describe the state-of-the-art in image detection. Then, a dataset of illustrations was collected from the Pixiv website using web scraping techniques. After data cleaning, separation, and augmentation, three pre-trained models were created and compared on 1200 training data and evaluated against 400 testing and 400 validation data. From the evaluation, the model using MobileNet V3 Large architecture achieved an impressive 94% accuracy, outperforming MobileNet V2 and Inception V3 architectures, respectively by 3% and 5%. Thus, the implementation of CNN holds the promise of providing an efficient solution for identifying and classifying various types of AI anime illustrations, benefiting consumers and artists practically. Future research could consider incorporating additional data categories and variations to further enhance the model's ability to distinguish between AI-generated and human-made illustrations.
Co-Authors Haezer, Eben Johan, Monika Evelin Wijaya, Westley Wong, Richard Faustine