Perceptual quality metric with internal generative mechanism

Area of Science:

• Computer Vision
• Image Processing
• Perceptual Science

Background:

• Objective image quality assessment (IQA) aims to mimic human perception but existing metrics struggle with diverse distortion types.
• Structural similarity metrics excel at content-dependent distortions, while peak signal-to-noise ratio (PSNR) is better for content-independent ones.
• The internal generative mechanism (IGM) theory suggests the human visual system actively predicts and minimizes uncertainty in image perception.

Purpose of the Study:

• To develop a novel IQA metric that accurately assesses image quality across various distortion types by integrating existing metric strengths.
• To leverage the internal generative mechanism (IGM) theory to guide the development of a more perceptually relevant IQA approach.
• To improve the consistency and accuracy of objective image quality evaluation compared to current state-of-the-art methods.

Main Methods:

• An autoregressive prediction algorithm is employed to decompose input images into predicted and disorderly portions based on IGM principles.
• Structural similarity is used to measure degradation in the predicted portion, representing primary visual information.
• Peak signal-to-noise ratio (PSNR) is applied to the disorderly portion, representing residual and uncertain information, with results combined based on noise energy.

Main Results:

• The proposed IQA metric demonstrates comparable performance to existing state-of-the-art methods across six public image databases.
• The decomposition strategy effectively addresses different distortion characteristics, leading to more accurate quality predictions.
• Experimental validation confirms the metric's ability to capture perceptual quality nuances influenced by various image degradations.

Conclusions:

• The proposed IQA metric, guided by IGM theory, offers a robust and accurate approach to evaluating image quality.
• Integrating structural similarity and PSNR based on image content decomposition enhances perceptual relevance.
• This method provides a significant advancement in objective image quality assessment, particularly for diverse and complex distortions.