No-Reference Quality Assessment of Tone-Mapped HDR Pictures.

Area of Science:

• Computer Vision
• Image Processing
• Perceptual Quality Assessment

Background:

• Automatic digital picture quality prediction is crucial for human visual experience.
• High Dynamic Range (HDR) images offer greater luminance and color depth but require tonemapping for standard displays.
• Existing no-reference image quality assessment (NR IQA) models struggle with tonemapped HDR content.

Purpose of the Study:

• To develop a novel no-reference image quality assessment (NR IQA) model specifically for High Dynamic Range (HDR) pictures.
• To improve the prediction accuracy of subjective image quality for HDR content.
• To create an algorithm, the HDR IMAGE GRADient-based Evaluator (HIGRADE), for assessing HDR image quality.

Main Methods:

• Developed NR IQA models using standard bandpass measurements and novel differential natural scene statistics (NSS) of HDR pictures.
• Incorporated both standard space-domain NSS features and new HDR-specific gradient-based features.
• Validated the models on a large-scale, crowdsourced HDR image database.

Main Results:

• The proposed HDR NR IQA models significantly elevate prediction performance compared to existing methods.
• The HIGRADE algorithm demonstrates superior accuracy in assessing the quality of HDR images.
• The models also show good performance on legacy Standard Dynamic Range (SDR) images.

Conclusions:

• The developed HDR NR IQA models and HIGRADE algorithm effectively predict perceived image quality for HDR content.
• The novel gradient-based features are key to improving performance on HDR images.
• The models offer a valuable tool for image quality assessment across both HDR and SDR formats.