Blind Quality Enhancement of Compressed Video via Fine-Grained Degradation-Guided Sequential Inference

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Video Processing & Computer Vision · Depth: Expert, extended

Summary

A new blind quality enhancement framework for compressed video (QECV) addresses limitations of existing methods by introducing a Degradation-aware Hierarchical Termination approach. This framework utilizes a Degradation Representation Learning (DRL) module, which employs dual-supervision with contrastive and classification learning, to extract fine-grained, multi-scale degradation representations from video content. Unlike prior blind methods that use global degradation vectors, DRL provides spatially detailed guidance for artifact removal. Additionally, a hierarchical termination mechanism dynamically adjusts the number of artifact reduction stages based on the video's compression level, optimizing computational efficiency. Experimental results on the MFQEv2 dataset, using HEVC and VVC standards, show significant performance gains. The method achieves a 110% PSNR improvement (from 0.31 dB to 0.65 dB) over a state-of-the-art blind method at QP = 22 and reduces average inference time at QP = 22 by half compared to QP = 42. It also demonstrates superior generalization to unseen quantization parameters and reduces computational complexity by approximately 73.7% compared to FBCNN.

Key takeaway

For Machine Learning Engineers developing video processing pipelines, especially for streaming or surveillance where quantization parameters are often unknown, this degradation-aware framework offers a compelling solution. You can achieve a 110% PSNR improvement at QP=22 and halve inference time for lighter compressions, significantly enhancing video quality and reducing computational overhead. Consider adopting this approach to deploy a single, adaptive model that maintains high visual fidelity across diverse compression levels. This will improve user experience and operational efficiency.

Key insights

Fine-grained, multi-scale degradation representation and adaptive processing stages significantly improve blind compressed video quality enhancement.

Principles

• Decoupling degradation from content enhances artifact removal.
• Dynamically adjusting processing stages optimizes computational efficiency.
• Fusing global and local features improves spatio-temporal dependency modeling.

Method

A pretrained Degradation Representation Learning (DRL) module extracts multi-scale degradation features using contrastive and classification learning. This guides a blind QECV network with a hierarchical termination mechanism and a dual-branch artifact reduction structure.

In practice

• Employ dual-supervision for robust degradation feature learning.
• Vary artifact reduction stages based on compression severity.
• Combine Transformer and dilated convolutions for spatio-temporal modeling.

Topics

• Blind Video Enhancement
• Degradation Representation Learning
• Hierarchical Termination
• Video Compression Artifacts
• HEVC
• VVC

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• DiffCVE: Diffusion-based Compressed Video Enhancement — DiffCVE enhances compressed video quality by integrating coding priors and compression awareness into a diffusion model.
• Learned JPEG Compression for DNN Vision — J4D optimizes JPEG for DNNs using a differentiable codec and rate estimator, significantly boosting accuracy or reducing file size.
• Learning Flexible Generalization in Video Quality Assessment by Bringing Device and Viewing Condition Distributions — Accurate video quality assessment on mobile devices requires incorporating device and viewing condition data.
• What Should a Streaming Video Model Remember? — SelectStream selectively allocates latent memory for streaming video understanding, balancing current perception with historical context.
• EvTexture++: Event-Driven Texture Enhancement for Video Super-Resolution — Event signals offer high-frequency details crucial for texture restoration and temporal consistency in video super-resolution.
• Exploiting Semantic and Pixel Representations for Ultra-Low Bitrate Image Compression — Ultra-low bitrate image compression can achieve pixel-level fidelity by combining semantic and pixel representations in a diffusion model.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.