Scholarlink

<p>Classical Image Quality Metrics (IQMs) are widely used in fields like remote sensing and computer vision for quantitative image discrimination. They face significant robustness challenges that compromise reliability in critical applications. These issues are evident when perceptually distinct images, especially after enhancements such as Spline-based non-linear Contrast Sketching, yield identical IQM scores, hindering effective differentiation and raising concerns about the trustworthiness of automated systems. To address these limitations, this study first analyzes the image discrimination robustness of 15 conventional IQMs on a dataset of 12 <i>Test Image</i>s, establishing baselines and failure points. Furthermore, this study presents and evaluates the <i>Localized IQM Calculation Method</i> (<i>Localized-IQM</i>), a potential enhancement to existing approaches. This method enhances IQM robustness by incorporating local image characteristics and statistical variations, moving beyond less discriminative global evaluations. Experimental findings demonstrate that <i>Localized-IQM</i> yields substantially more robust and reliable outcomes for challenging image discrimination problems compared to the direct global application of conventional IQMs. Its enhanced performance is consistent across diverse image types and common alteration levels. Consequently, the proposed method offers a significantly improved, dependable framework for image quality assessment, promising to advance image analysis accuracy and reliability in fields requiring precise image differentiation.</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

Image discrimination robustness of classical image quality metrics: an analysis on MAE, MSE, ERGAS, LMSE, SSIM, SAM, UIQI, PSNR, NIQE, PIQE, SNR, VIF, FSIM, GMSD, and NCC

  • 期刊论文

  • 发表日期：

    2025年10月4日

查看全文
  • Pinar Civicioglu,
  • Erkan Besdok

摘要

Classical Image Quality Metrics (IQMs) are widely used in fields like remote sensing and computer vision for quantitative image discrimination. They face significant robustness challenges that compromise reliability in critical applications. These issues are evident when perceptually distinct images, especially after enhancements such as Spline-based non-linear Contrast Sketching, yield identical IQM scores, hindering effective differentiation and raising concerns about the trustworthiness of automated systems. To address these limitations, this study first analyzes the image discrimination robustness of 15 conventional IQMs on a dataset of 12 Test Images, establishing baselines and failure points. Furthermore, this study presents and evaluates the Localized IQM Calculation Method (Localized-IQM), a potential enhancement to existing approaches. This method enhances IQM robustness by incorporating local image characteristics and statistical variations, moving beyond less discriminative global evaluations. Experimental findings demonstrate that Localized-IQM yields substantially more robust and reliable outcomes for challenging image discrimination problems compared to the direct global application of conventional IQMs. Its enhanced performance is consistent across diverse image types and common alteration levels. Consequently, the proposed method offers a significantly improved, dependable framework for image quality assessment, promising to advance image analysis accuracy and reliability in fields requiring precise image differentiation.