Enhancing Vehicle License Plate Image Quality Using Median Filtering: Implications for ANPR Performance

Abstract

Automatic vehicle license plate detection is a crucial component in traffic monitoring and law enforcement. One of the primary challenges in Automatic Number Plate Recognition (ANPR) systems is the presence of impulsive noise, such as salt-and-pepper noise, which can disrupt the accuracy of character detection. This study serves as a baseline for ANPR research. The objective of this study is to optimize license plate detection by applying the Median Filtering method, while simultaneously comparing its performance on white and black license plates. Evaluation was conducted using 100 samples of Indonesian vehicle license plates, consisting of 50 white plates and 50 black plates, under consistent lighting conditions and angles. The testing results indicate that for white plates, the Mean Squared Error (MSE) ranged from 0.000018 to 0.000564, with an average Peak Signal-to-Noise Ratio (PSNR) of 38.38 dB and an average processing time of 1.105 seconds. For black plates, the MSE ranged from 0.000028 to 0.000592, with an average PSNR of 37.82 dB and an average processing time of 1.150 seconds. Comparative analysis shows that white plates produced slightly higher image quality than black plates, although both types of plates were processed efficiently. Overall, the Median Filtering method proved effective in reducing impulsive noise, preserving the sharpness of license plate characters, and maintaining processing speed. These findings demonstrate that Median Filtering is a reliable method for enhancing the accuracy and efficiency of ANPR systems under various license plate colors and lighting conditions.