Scholarlink

The presence of geometric distortions in real-world images is pervasive and arises from various sources, including camera misalignment, viewpoint changes, and lens imperfections. These distortions can lead to misalignment of objects, making it challenging for deep learning models to accurately identify and classify them. Geometric correction is a foundational task in computer vision, aimed at rectifying spatial distortions and misalignments within images. This paper presents a novel CNN + STN model tailored for geometric correction and explores its efficacy in rectifying spatial distortions and misalignments within images. Through extensive experimentation, CNN + STN model’s performance is rigorously evaluated on synthetic dataset. Experimental results reveal its ability to enhance image quality, accurately register visual data, and improve the performance of downstream computer vision tasks. The integration of CNN + STN promises to unlock new frontiers in computer vision applications, solidifying its position as an indispensable asset in the computer vision toolkit.

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

Enhancing Image Quality: Geometric Correction with CNN + STN Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • J. Aswini,
  • A. Revathi,
  • Ramayanam Jhansi Rani,
  • Kuntrapakam Divya,
  • T. Ravi Kumar,
  • A. Basi Reddy

摘要

The presence of geometric distortions in real-world images is pervasive and arises from various sources, including camera misalignment, viewpoint changes, and lens imperfections. These distortions can lead to misalignment of objects, making it challenging for deep learning models to accurately identify and classify them. Geometric correction is a foundational task in computer vision, aimed at rectifying spatial distortions and misalignments within images. This paper presents a novel CNN + STN model tailored for geometric correction and explores its efficacy in rectifying spatial distortions and misalignments within images. Through extensive experimentation, CNN + STN model’s performance is rigorously evaluated on synthetic dataset. Experimental results reveal its ability to enhance image quality, accurately register visual data, and improve the performance of downstream computer vision tasks. The integration of CNN + STN promises to unlock new frontiers in computer vision applications, solidifying its position as an indispensable asset in the computer vision toolkit.