License Plate Recognition System Using Traditional Image Processing and OCR on KRIA KR260 (CPU-Based Implementation)
Year: 2026
Type: conference-paper
Abstract: In this paper, we present a lightweight and highspeed LPR system which is designed only to run on the CPU of the KRIA KR260 platform. Most existing solutions depend on deep learning models with thousands of parameters but our system utilizes the combination of classical image processing algorithms and Tesseract OCR to provide good performance for recognition with efficient computation. The image processing consists of several steps including gray scale conversion, adaptive thresholding, edge detection, and contour based LPR. All the processing stages are tuned for optimal performance of the following OCR operations. Additionally, we apply regex based filtering for checking the recognized results. This way can minimize false positives and increase the overall system reliability. Through the experiments, the LPR detection rate is up to 94 percent and the average processing speed per frame is 0.60 s. It outperforms non-optimized solution. Comparing to cloud-based deep learning based OCR (Easy OCR), it gives the robust results but costs high computing resource and long delay. Running Easy OCR on KRIA will encounter significant difficulties on dependency and computing resource. A significant percentage of false recognize outputs (mostly from character confusion and noisy image) have been eliminated by adjusting preprocessing and checking method.
Institution: Chaitanya Bharathi Institute of Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7203855853