Deblurring Anime Faces using Deep Convolutional Neural Networks

• You can find the dataset here on Kaggle.

• 1,000 images of the dataset is already in the animefaces folder, to use the whole dataset extract it inside this folder

The directory structure for this project:

├───input
│   ├───animefaces
│   ├───box_filter_blurred
│   ├───gaussian_blurred
│   |───greyscaled
│   └───motion_blurred
├───outputs
│   ├───box_filter_deblurred
│   ├───gaussian_deblurred
│   └───motion_deblurred
└───src

How to Execute

In src example using box blur

1. grayscale.py
2. box_filter_blur.py
3. train_box_filter.py

Example Result Images

The outputs are sorted in this order for each section (left to right, downwards):

• Original image
• Grayscaled image
• Blurred image
• Model 1 deblurred image
• Model 2 deblurred image
• Model 3 deblurred image

Box Filter Deblurring

Learning Rate = 10^-4:

More on our website for this project here

References

1. C. Dong, C. C. Loy, K. He and X. Tang, "Image Super-Resolution Using Deep Convolutional Networks," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 38, no. 2, pp. 295-307, 1 Feb. 2016, doi: 10.1109/TPAMI.2015.2439281.
2. Jian Sun, Wenfei Cao, Zongben Xu, Jean Ponce. Learning a convolutional neural network for non-uniform motion blur removal. CVPR 2015 - IEEE Conference on Computer Vision and Pattern Recognition 2015, Jun 2015, Boston, United States. IEEE, 2015,.
3. Ledig, Christian & Theis, Lucas & Huszar, Ferenc & Caballero, Jose & Cunningham, Andrew & Acosta, Alejandro & Aitken, Andrew & Tejani, Alykhan & Totz, Johannes & Wang, Zehan & Shi, Wenzhe. (2017). Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network. 105-114. 10.1109/CVPR.2017.19.
4. Albluwi, Fatma & Krylov, Vladimir A. & Dahyot, Rozenn. (2018). Image Deblurring and Super-Resolution Using Deep Convolutional Neural Networks. 1-6. 10.1109/MLSP.2018.8516983.
5. Kupyn, Orest & Budzan, Volodymyr & Mykhailych, Mykola & Mishkin, Dmytro & Matas, Jiri. (2017). DeblurGAN: Blind Motion Deblurring Using Conditional Adversarial Networks.