Abstract:Being captured by amateur photographers, reciprocally propagated through multimedia pipelines, and compressed with different levels, real-world images usually suffer from a wide variety of hybrid distortions. Faced with this scenario, full-reference (FR) image quality assessment (IQA) algorithms can not deliver promising predictions due to the inferior references. Meanwhile, existing no-reference (NR) IQA algorithms remain limited in their efficacy to deal with different distortion types. To address this obstacle, we explore a NR-IQA metric by predicting the perceptual quality of distorted-then-compressed images using a deep neural network (DNN). First, we propose a novel two-stream DNN to handle both authentic distortions and synthetic compressions and adopt effective strategies to pre-train the two branches of the network. Specifically, we transfer the knowledge learned from in-the-wild images to account for authentic distortions by utilizing a pre-trained deep convolutional neural network (CNN) to provide meaningful initializations. Meanwhile, we build a CNN for synthetic compressions and pre-train it on a dataset including synthetic compressed images. Subsequently, we bilinearly pool these two sets of features as the image representation. The overall network is fine-tuned on an elaborately-designed auxiliary dataset, which is annotated by a reliable objective quality metric. Furthermore, we integrate the output of the authentic-distortion-aware branch with that of the overall network following a two-step prediction manner to boost the prediction performance, which can be applied in the distorted-then-compressed scenario when the reference image is available. Extensive experimental results on several databases especially on the LIVE Wild Compressed Picture Quality Database show that the proposed method achieves state-of-the-art performance with good generalizability and moderate computational complexity.

In-training Restoration Models Matter: Data Augmentation for Degraded-reference Image Quality Assessment

In-training Restoration Models Matter

Joint Distortion Restoration and Quality Feature Learning for No-Reference Image Quality Assessment

Learning Conditional Knowledge Distillation for Degraded-Reference Image Quality Assessment

Image Quality Assessment–driven Reinforcement Learning for Mixed Distorted Image Restoration

Hierarchical Curriculum Learning for No-Reference Image Quality Assessment

Contrastive Distortion‐level Learning‐based No‐reference Image‐quality Assessment

Relationship Quantification of Image Degradations

Learning to Predict the Quality of Distorted-then-compressed Images Via a Deep Neural Network

Image Quality Assessment: Enhancing Perceptual Exploration and Interpretation with Collaborative Feature Refinement and Hausdorff distance

Synthesizing Realistic Image Restoration Training Pairs: A Diffusion Approach

Rdm-Ir: Task-Adaptive Deep Unfolding Network for All-in-One Image Restoration

Learning degradation priors for reliable no-reference image quality assessment

Feature Rectification and Enhancement for No-Reference Image Quality Assessment

Conformer and Blind Noisy Students for Improved Image Quality Assessment

Diffusion Model Based Visual Compensation Guidance and Visual Difference Analysis for No-Reference Image Quality Assessment

Reference-based Multi-stage Progressive Restoration for Multi-degraded Images

RAN4IQA: Restorative Adversarial Nets for No-Reference Image Quality Assessment

No-Reference Image Quality Assessment: an Attention Driven Approach

A Full-Reference Image Quality Assessment Method Via Deep Meta-Learning and Conformer

DRM-IR: Task-Adaptive Deep Unfolding Network for All-In-One Image Restoration