양태원 나건열 2024-01-19T08:02:37Z 2024-01-19T08:02:37Z 2024-01-03 2023-12 2508-4003 https://pubs.kist.re.kr/handle/201004/113024 In vehicular data, license plates inherently contain personally identifiable information, prompting escalating privacy concerns. Nonetheless, recognizing these plates is vital for applications like traffic surveillance and automated toll collection. With advanced deep learning paradigms, notably convolutional neural networks, license plate recognition has achieved remarkable precision, outperforming traditional optical character recognition. In this study, the authors present a methodology for identifying license plates from comprehensive vehicle images and replacing them with synthesized versions using advanced image processing and deep learning. A dataset of 1,217 vehicle images, representing 11 license plate categories, was curated. Using this dataset, license plate images, after object detection, undergo a sequential process integrating a plate region delineation model and a plate classification model. Synthesized license plates, derived from the model's output, are then overlaid onto the original vehicle images. This methodology results in a license plate alteration system compatible with 11 diverse plate categories. Korean 한국CDE학회 EfficientNetV2를 활용한 차량 번호판 비식별화 시스템 개발 Development of a Vehicle License Plate De-identification System using EfficientNetV2 Article 10.7315/CDE.2023.503 2 한국CDE학회 논문집, v.28, no.4, pp.503 - 513 한국CDE학회 논문집 28 4 503 513 N kci ART003020357 Deep learning De-identification License plate recognition Object recognition