Scholarlink

<p>The limited dynamic range of digital imaging sensors often leads to under- or over-exposed images. While deep learning methods currently dominate multi-exposure image fusion (MEF), they suffer from data dependency and poor interpretability. This paper proposes a novel model-based MEF framework using the Discrete Band-Limited Shearlet Transform (DBLST), which provides superior directional representation compared to traditional wavelets. Our method decomposes source images using DBLST and fuses coefficients according to specifically designed rules for low-frequency and high-frequency components. Extensive experiments demonstrate that the proposed algorithm achieves superior performance to several representative transform-based methods (DWT, NSCT, NSST) in terms of detail preservation and information richness, positioning DBLST as a powerful tool for model-based fusion. The results confirm DBLST as an efficient and interpretable, training-free alternative to data-driven deep learning models for rendering high-dynamic-range images, particularly in scenarios where transparency and reproducibility are prioritized.</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

A model-based image fusion framework using discrete band-limited shearlets

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月24日

查看全文
  • Wentao Ji,
  • Xing Chen

摘要

The limited dynamic range of digital imaging sensors often leads to under- or over-exposed images. While deep learning methods currently dominate multi-exposure image fusion (MEF), they suffer from data dependency and poor interpretability. This paper proposes a novel model-based MEF framework using the Discrete Band-Limited Shearlet Transform (DBLST), which provides superior directional representation compared to traditional wavelets. Our method decomposes source images using DBLST and fuses coefficients according to specifically designed rules for low-frequency and high-frequency components. Extensive experiments demonstrate that the proposed algorithm achieves superior performance to several representative transform-based methods (DWT, NSCT, NSST) in terms of detail preservation and information richness, positioning DBLST as a powerful tool for model-based fusion. The results confirm DBLST as an efficient and interpretable, training-free alternative to data-driven deep learning models for rendering high-dynamic-range images, particularly in scenarios where transparency and reproducibility are prioritized.