Significant Improvement in License Plate Recognition through Image Deduplication

Abstract

The detection of license plates, or Automatic Number Plate Recognition (ANPR), is crucial for applications in parking management, vehicle tracking, and security. However, the efficiency of ANPR systems is often compromised by large datasets containing numerous similar or duplicate images, leading to increased storage costs and slowed processing times. This research proposes an innovative approach that combines perceptual hashing and locality-sensitive hashing (LSH) to enhance the detection of redundant images. Perceptual hashing generates unique visual fingerprints for images, facilitating efficient duplicate identification, while LSH groups similar images to reduce false positives. Additionally, Optical Character Recognition (OCR) is applied to the image pairs identified by LSH to extract license plates and verify vehicle identity. By integrating these techniques, the proposed method effectively mitigates redundancy, optimizing storage and improving the performance of ANPR systems for accurate real-world recognition.