Semi-Supervised Blind Quality Assessment with Confidence-quantifiable Pseudo-label Learning for Authentic Images

Yan Zhong, Chenxi Yang, Suyuan Zhao, Tingting Jiang

TL;DR: This paper introduces a novel semi-supervised framework for blind image quality assessment based on label propagation, utilizing confidence-quantifiable pseudo-labels to enhance performance.

Abstract: This paper presents CPL-IQA, a novel semi-supervised blind image quality assessment (BIQA) framework for authentic distortion scenarios. To address the challenge of limited labeled data in IQA area, our approach leverages confidence-quantifiable pseudo-label learning to effectively utilize unlabeled authentically distorted images. The framework operates through a preprocessing stage and two training phases: first converting MOS labels to vector labels via entropy minimization, followed by an iterative process that alternates between model training and label optimization. The key innovations of CPL-IQA include a manifold assumption-based label optimization strategy and a confidence learning method for pseudo-labels, which enhance reliability and mitigate outlier effects. Experimental results demonstrate the framework's superior performance on real-world distorted image datasets, offering a more standardized semi-supervised learning paradigm without requiring additional supervision or network complexity.

Lay Summary:

Assessing the quality of images is crucial for applications like photography, social media, and medical imaging. However, most methods require large amounts of labeled data, which is expensive and time-consuming to collect. This is especially challenging for "authentic" images—real-world photos with natural distortions like blur or noise—since their quality scores often vary even among similar-looking images.

We propose a new method called CPL-IQA that combines labeled and unlabeled images to train a quality assessment model more efficiently. Our key innovation is a technique to estimate confidence scores for "pseudo-labels" (predicted quality scores for unlabeled images). By converting simple quality score labels into vector labels and refining them iteratively, our method ensures reliable predictions without needing extra data or complex networks.

CPL-IQA outperforms existing methods on real-world image datasets, offering a practical solution for scenarios where labeled data is scarce. This could benefit industries relying on image quality, from smartphone cameras to healthcare imaging, by reducing the need for costly manual labeling while maintaining accuracy.

Primary Area: Applications->Computer Vision

Keywords: BIQA, Semi-supervised Learning, Pseudo-labels, Confidence Learning, Real-world Distorted Images

Submission Number: 1461