Detailgrad

Detailgrad is a quantitative metric proposed for digital image quality assessment to quantify how well fine details are preserved in a processed image relative to a high-quality reference image. The measure concentrates on local gradient information, arguing that edges and textures are primary indicators of perceived detail.

Computational approach: The metric operates on a grayscale version or luminance channel. It computes gradient magnitude maps using a gradient operator such as Sobel or Scharr. The difference between the gradient maps of the reference and processed images is measured, for example by mean absolute error or a robust error function, and aggregated over all pixels. The resulting score can be normalized to the range [0,1] or [0,100], with higher values indicating better preservation of detail.

History and name: The term Detailgrad combines 'detail' and 'gradient' and has appeared in peer-reviewed studies since the 2010s as part of efforts to supplement traditional metrics with perceptually relevant detail measures. It is presented as a concept within image quality assessment rather than a universally standardized metric.

Applications: It is used to evaluate image compression, denoising, restoration, and super-resolution algorithms, often alongside SSIM or VIF. It helps distinguish methods that preserve texture from those that blur edges.

Limitations: It depends on a reference image and the chosen gradient operator; sensitive to misalignment, illumination changes, and color-to-grayscale conversion; may overemphasize small high-frequency artifacts. It is typically used as a complementary metric rather than a standalone quality score.