Learning Single-Image Super-Resolution in the JPEG Compressed Domain

2025-12-03

AI Summary

In JPEG-compressed-domain single-image super-resolution (SISR), data loading poses a critical bottleneck for both training and inference. To address this, we propose a lightweight deep network that operates directly in the DCT coefficient domain. Unlike conventional approaches requiring full JPEG decoding into the pixel domain, our method is the first to use quantized DCT coefficients—the native JPEG-encoded features—as direct network input, thereby bypassing pixel-domain reconstruction and significantly reducing I/O and computational overhead. The architecture incorporates frequency-domain attention and block-wise adaptive reconstruction modules, enabling end-to-end compressed-domain SISR while preserving visual fidelity. Experiments demonstrate a 2.6× speedup in data loading, a 2.5× acceleration in training, and competitive PSNR/SSIM performance relative to state-of-the-art pixel-domain methods. This work establishes a new paradigm for efficient edge deployment of SISR.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Learning & Optimization for CV
• Data Mining & Knowledge Management: Data Compression

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Deep learning models have grown increasingly complex, with input data sizes scaling accordingly. Despite substantial advances in specialized deep learning hardware, data loading continues to be a major bottleneck that limits training and inference speed. To address this challenge, we propose training models directly on encoded JPEG features, reducing the computational overhead associated with full JPEG decoding and significantly improving data loading efficiency. While prior works have focused on recognition tasks, we investigate the effectiveness of this approach for the restoration task of single-image super-resolution (SISR). We present a lightweight super-resolution pipeline that operates on JPEG discrete cosine transform (DCT) coefficients in the frequency domain. Our pipeline achieves a 2.6x speedup in data loading and a 2.5x speedup in training, while preserving visual quality comparable to standard SISR approaches.

Problem

• Training models directly on JPEG features to reduce decoding overhead
• Improving data loading efficiency for super-resolution tasks
• Operating on JPEG DCT coefficients to speed up training

Innovation

• Training models directly on JPEG encoded features
• Lightweight super-resolution pipeline using DCT coefficients
• Achieves significant speedup in data loading and training