An Unsupervised Information-Theoretic Perceptual Quality Metric
This work addresses the problem of costly supervised data for image quality assessment by developing an unsupervised metric, offering a practical solution for researchers and practitioners in computer vision.
The paper tackles the challenge of building tractable models of human perception for image quality assessment by introducing the Perceptual Information Metric (PIM), which combines information-theoretic objectives with a human visual system-inspired architecture and unsupervised training on video frames. It shows that PIM is competitive with supervised metrics on the BAPPS dataset and outperforms them in ranking image compression methods on CLIC 2020.
Tractable models of human perception have proved to be challenging to build. Hand-designed models such as MS-SSIM remain popular predictors of human image quality judgements due to their simplicity and speed. Recent modern deep learning approaches can perform better, but they rely on supervised data which can be costly to gather: large sets of class labels such as ImageNet, image quality ratings, or both. We combine recent advances in information-theoretic objective functions with a computational architecture informed by the physiology of the human visual system and unsupervised training on pairs of video frames, yielding our Perceptual Information Metric (PIM). We show that PIM is competitive with supervised metrics on the recent and challenging BAPPS image quality assessment dataset and outperforms them in predicting the ranking of image compression methods in CLIC 2020. We also perform qualitative experiments using the ImageNet-C dataset, and establish that PIM is robust with respect to architectural details.