Natural Image Quality Evaluator (NIQE)

Natural image quality evaluator (NIQE) calculates the no-reference image quality score for images that may be distorted or of low perceptual quality. The contribution of this metric is derived from not requiring a known class of image distortions or perception degradations since the measure is not produced with manually-degraded data. Theoretically, this would allow the measure to be more robust to unforeseeable image quality issues. A smaller score indicates better perceptual quality.

NIQE can support the Safety objective in contexts where poor image quality could lead to harmful decisions or outputs, such as in medical diagnostics or autonomous navigation. By quantifying image quality, NIQE helps ensure that AI systems operate on reliable visual data, thereby reducing the risk of harm due to degraded or distorted images. However, this connection is indirect and context-dependent, and NIQE does not address other Trustworthy AI objectives.

Trustworthy AI Relevance

This metric addresses Robustness and Human Agency & Control by quantifying relevant system properties. Robustness: NIQE quantifies perceptual quality and naturalness of images without a reference image, enabling detection of degradation, noise, compression artifacts, or distribution shifts in visual outputs. As such it supports assessing an AI system's resilience to adverse conditions (e.g., noisy inputs, model degradation, OOD visual artifacts) and can be used in monitoring/validation pipelines to maintain consistent output quality. Human Agency & Control: NIQE provides an objective, automatically computed quality score that can be surfaced to users or used to gate outputs (e.g., flag or hide low-quality/generated images, trigger human review).