DCT-RCAN

DCT-Based Residual Network for NIR Image Colorization

[📄 Paper Link (IEEE Xplore)]

Authors:
Hongcheng Jiang,
Paras Maharjan,
Zhu Li,
George York

---

🔍 Overview

This repository provides the official implementation of
"DCT-Based Residual Network for NIR Image Colorization",
Published in IEEE ICIP 2022.

📈 Performance Gains

We propose DCT-RCAN, a DCT-guided residual network for NIR image colorization.

✅ Results on Validation Set

• PSNR: 22.15 dB (+1.48 dB, +7.2%)
• SSIM: 0.77 (+0.09, +13.2%)
• AE: 3.40° (–0.57°, –14.4%)

---

🧠 Network Architecture

Network Architecture of DCT-RCAN

---

📊 Quantitative Results

Table: Average PSNR (dB), SSIM, and Angular Error (AE in degrees) on the validation dataset.

Model	PSNR (dB)	SSIM	AE (°)
MFF	17.39	0.61	4.69
ATcycleGAN	20.67	0.68	3.97
SST	14.26	0.57	5.61
SPADE	19.24	0.59	4.59
NIR-GNN	17.50	0.60	5.22
Proposed Method	22.15	0.77	3.40

---

🖼️ Visual Results

2D-DCT Visualization from a NIR Image

Comparison with State-of-the-Art Methods

---

📬 Contact

If you have any questions, feedback, or collaboration ideas, feel free to reach out:

• 💻 Website: jianghongcheng.github.io
• 📧 Email: [email protected]
• 🏫 Affiliation: University of Missouri–Kansas City (UMKC)

---

📚 Citation

If you find this work helpful in your research, please cite:

@inproceedings{jiang2022dct,
  title={DCT-Based Residual Network for NIR Image Colorization},
  author={Jiang, Hongcheng and Maharjan, Paras and Li, Zhu and York, George},
  booktitle={2022 IEEE International Conference on Image Processing (ICIP)},
  pages={2926--2930},
  year={2022},
  organization={IEEE}
}