Denoising Autoencoder-based Feature Compression for Enhanced Performance of Vehicle License Plate Image Recognition

To combat the issue of low-quality license plate images captured by CCTV cameras in Electronic Traffic Law Enforcement (ETLE) systems, this study presents a Denoising Autoencoder-based feature compression method for improved image recognition. The Denoising Autoencoder significantly outperforms the conventional Autoencoder, enhancing feature extraction by introducing noise to the input. The method increases the accuracy rate from the standalone use of a Convolutional Neural Network $(90.47 \%$) to $94.76 \%$, showcasing its effectiveness in noise reduction, dimensionality reduction, and feature learning from license plate images.