Seungju Lee is a researcher specializing in computer vision and deep learning, with a particular focus on optical character recognition (OCR) and real-world intelligent vision systems. His recent work centers on a systematic comparative study of OCR architectures for Korean license plate recognition, where he investigates the practical limitations and deployment-oriented trade-offs of different sequence modeling strategies under controlled experimental conditions. His research examines two representative OCR paradigms: CNN-based feature extraction with Attention-LSTM sequence modeling and lightweight MobileNetV3 backbones combined with Transformer decoders. By fixing the detection stage using a unified YOLO-based framework and conducting controlled ablation experiments with a shared backbone, he analyzes whether performance differences arise primarily from sequence decoders or from feature representation quality. Through multi-dataset evaluation and cross-platform experiments—including GPU and embedded devices—his work provides quantitative insights into accuracy, computational complexity, and real-time deployment feasibility. A key contribution of his research is the identification and quantitative validation of tracking-induced region-of-interest (ROI) instability as a dominant factor affecting recognition performance in sequential license plate recognition scenarios. His analysis demonstrates that cumulative spatial drift, jitter, and frozen ROI propagation errors often