|
Original research INVERSE FILTER-BASED IMAGE RECONSTRUCTION FOR IMPROVED ANPR PERFORMANCE IN BANGLADESH SECURITY SYSTEMSPages 111-120
Abstract
This paper presents a novel inverse filter-based image reconstruction framework tailored to enhance Automatic Number Plate Recognition (ANPR) performance under the challenging environmental conditions prevalent in Bangladesh. While prior systems leveraging YOLOv8n detection and CRNN recognition have shown promise, their performance deteriorates in the presence of rain, dust, motion blur, and low-light conditions. To address these issues, the proposed method introduces a pre-processing pipeline employing frequency-domain inverse filtering with region-specific degradation models, including monsoon-induced blur and particulate haze. The system integrates seamlessly with existing ANPR infrastructure, adding only 23 ms latency while achieving a significant performance boost notably a 96.8% mAP@0.5 under rainy conditions, an 8.1% improvement over the baseline. Real-world evaluations across Dhaka, Chattogram, and Sylhet validate the system’s robustness, with a 33% reduction in character segmentation errors for dust-affected plates. This research confirms that targeted image reconstruction is vital for the deployment of resilient ANPR systems in environmentally constrained and infrastructure-limited regions.
Keywords: Inverse Filtering, ANPR, Image Reconstruction, Environmental Degradation, Frequency-Domain Restoration, Real-Time Processing.
|
Asif Eakball Emon, 
PDF 