Scholarlink

<p>In the domain of deep blind image deblurring, inaccurate estimation of the blur kernel size often leads to noticeable artifacts in the deblurred output. This paper introduces a novel method for automatically estimating the optimal blur kernel size directly from a given blurry image. By leveraging Histogram of Oriented Gradients (HOG) features and Support Vector Regression (SVR) models, our approach eliminates the need for assumptions about the blur kernel type. Experimental results demonstrate that our method outperforms existing techniques in accurately estimating blur kernel sizes, thereby reducing deblurring artifacts. Furthermore, we enhance the deblurring performance by integrating Deep Image Prior (DIP) and Fully Connected Networks (FCNs) to capture clean image and blur kernel priors, respectively. This allows for joint optimization, representing a significant advancement in deep learning-based blind image deblurring. The proposed method achieves higher accuracy in blur kernel size estimation, faster processing times, and improved deblurred image quality, as evidenced by metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM).</p>

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

Automatic Estimation of Blur Kernel Size for Enhanced Deep Blind Image Deblurring

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Mitra Abdollahi,
  • Alireza Ahmadyfard

摘要

In the domain of deep blind image deblurring, inaccurate estimation of the blur kernel size often leads to noticeable artifacts in the deblurred output. This paper introduces a novel method for automatically estimating the optimal blur kernel size directly from a given blurry image. By leveraging Histogram of Oriented Gradients (HOG) features and Support Vector Regression (SVR) models, our approach eliminates the need for assumptions about the blur kernel type. Experimental results demonstrate that our method outperforms existing techniques in accurately estimating blur kernel sizes, thereby reducing deblurring artifacts. Furthermore, we enhance the deblurring performance by integrating Deep Image Prior (DIP) and Fully Connected Networks (FCNs) to capture clean image and blur kernel priors, respectively. This allows for joint optimization, representing a significant advancement in deep learning-based blind image deblurring. The proposed method achieves higher accuracy in blur kernel size estimation, faster processing times, and improved deblurred image quality, as evidenced by metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM).