Research map / Image generation / Image restoration
Uncover Treasures in DCT: Advancing JPEG Quality Enhancement by Exploiting Latent Correlations
Jing Yang, Qunliang Xing, Mai Xu, Minglang Qiao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add block-based convolution EMBC modules instead of ResBlocks (ablation) | AJQE w/ FBCNN · Color JPEG, QF10 | PSNR (dB): 27.29 -> 27.69 | Improved |
| add the point-based correlation module EMPC instead of a ResBlock (ablation) | AJQE w/ FBCNN · Color JPEG, QF10 | PSNR (dB): 27.59 -> 27.69 | Improved |
| use EMBC and EMPC together instead of ResBlocks for both (ablation) | AJQE w/ FBCNN · Color JPEG, QF10 | PSNR (dB): 26.92 -> 27.69 | Improved |
| use block-based convolution instead of vanilla convolution in EMBC (ablation) | AJQE w/ FBCNN · Color JPEG | PSNR (dB): 0.14 dB decrease in PSNR for BConv-VConv | Improved |
| use UMEP instead of spatial nearest-neighbor upsampling of chroma DCT (ablation) | AJQE w/ FBCNN · Color JPEG | PSNR (dB): average decrease in PSNR of 0.16 dB | Improved |
| use UMEP instead of transposed-convolution upsampling of chroma DCT (ablation) | AJQE w/ FBCNN · Color JPEG | PSNR (dB): average decrease in PSNR of 0.10 dB | Improved |
| use UMEP instead of spatial nearest-neighbor upsampling at QF40 (ablation) | AJQE w/ FBCNN · Color JPEG, QF40 | PSNR (dB): 32.24 -> 32.41 | Improved |
| adapt pixel-domain model to the DCT domain with AJQE | pixel-domain models · Color JPEG | PSNR (dB): 0.35 dB improvement in PSNR | Improved |
| adapt pixel-domain model to the DCT domain with AJQE | pixel-domain models · Grayscale JPEG | PSNR (dB): 0.39 dB improvement in PSNR | Improved |
| AJQE versus existing DCT-domain methods | QGAC and JDEC · Color JPEG | PSNR (dB): 0.41 dB improvement | Improved |
| AJQE versus pixel-domain models on efficiency | pixel-domain models · BSDS500 | enhancement throughput: 60.5% higher enhancement throughput | Improved |
| AJQE versus pixel-domain models on compute | pixel-domain models · BSDS500 | FLOPs: 45.9% fewer FLOPs | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- JDEC — JDEC: JPEG Decoding via Enhanced Continuous Cosine Coefficients (CVPR 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- JDEC: JPEG Decoding via Enhanced Continuous Cosine Coefficients (CVPR 2024 · similarity 0.87)
- DCTMamba: Advancing JPEG Image Restoration Through Long-Sequence Modeling and Adaptive Frequency Strategy (AAAI 2025 · similarity 0.87)
- Breaking the Multi-Enhancement Bottleneck: Domain-Consistent Quality Enhancement for Compressed Images (TPAMI 2025 · similarity 0.86)
- DCT-Shield: A Robust Frequency Domain Defense against Malicious Image Editing (ICCV 2025 · similarity 0.86)
- Enhancing Quality of Compressed Images by Mitigating Enhancement Bias Towards Compression Domain (CVPR 2024 · similarity 0.85)
- DCTdiff: Intriguing Properties of Image Generative Modeling in the DCT Space (ICML 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
