Scholarlink

Given the complexity of variations in CCTV (Closed Circuit Television) scenarios and equipment, it is essential to adopt advanced image enhancement methods in Automatic License Plate Recognition (ALPR). This study investigated the impact of two distinct image degradation methods during the synthesis of low-resolution images for constructing paired datasets, used in training models based on the Real-ESRGAN super-resolution architecture. The results demonstrated a significant improvement in generalization, both in the quality of the generated images and in the correct recognition rate of license plate characters. Training datasets created using a more robust degradation method, composed of multiple degradation stages, exhibited superior performance.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Study on the Impact of the Degradation Method on the Generalization of Super-Resolution Models for ALPR

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Cristiano L. Oliveira,
  • Leonardo N. Matos,
  • Paulo S. G. de Mattos Neto,
  • Paulo Novais,
  • Flávio A. O. Santos,
  • Marcelo H. L. Barreto

摘要

Given the complexity of variations in CCTV (Closed Circuit Television) scenarios and equipment, it is essential to adopt advanced image enhancement methods in Automatic License Plate Recognition (ALPR). This study investigated the impact of two distinct image degradation methods during the synthesis of low-resolution images for constructing paired datasets, used in training models based on the Real-ESRGAN super-resolution architecture. The results demonstrated a significant improvement in generalization, both in the quality of the generated images and in the correct recognition rate of license plate characters. Training datasets created using a more robust degradation method, composed of multiple degradation stages, exhibited superior performance.