DBSR

Super-Resolution on Degraded Low-Resolution Images Using Convolutional Neural Networks (DBSR)

DBSR Network

1. DBSRCNN Network: Image Deblurring And Super-Resolution Using Deep Convolutional Neural Networks, F. Albluwi, V. Krylov and R. Dahyot, IEEE International Workshop on Machine Learning for Signal Processing (MLSP 2018 http://mlsp2018.conwiz.dk/home.htm ), September 2018, Aalborg, Danemark.

2. DBSR Network: DBSR Network is an extension of DBSRCNN Network with extra 3 layers to enhance the extracted features inside the network.

DBSRCNN-Keras

This code is to process the blurred low-resolution images to get deblurred high-residual images.

If this code is helpful for you, please cite this paper: Super-Resolution on Degraded Low-ResolutionImages Using Convolutional Neural Networks, F. Albluwi, V. Krylov and R. Dahyot, 27th European Signal Processing Conference (Eusipco 2019 ), September 2019.

Dependencies

1. Python 3.6.5, and above.
2. TensorFlow 1.1.0, and above.
3. Keras 2.2.2, and above.
4. Matlab.
5. Matconvnet.

Generating data

1. blur images by Gaussian filter (imgaussfilt) at different levels (sigma = 1, 2, and 3).
2. resize images with 'bicubic' function using upscaling factor = 3, published papers recently generally use Matlab to produce low-resolution image.
3. For a fair comparison with SRCNN network; training set 291 images (Yang91 + 200 BSD) are used.

Training

1. Generate training patches using Matlab: run generate_train.m and generate_test.m.
2. Use Keras with TensorFlow (tf) as a backend to train DBSRCNN model; Adam is used to optimizing the network for fast convergence: run DBSRCNN_train.py to produce DBSRCNN_blur model.
3. Convert Keras model to.Mat for testing using Matconvnet: run load_save.py first, then run save_model.m to produce Matconvnet model.
4. Run NB_SRCNN_Concat_blur_test.m in “test” folder to test the model; Set5 and Set14 are used as testing data.

Some Qualitative Results

SISR with different models on images after Gaussian blur with different sigma = 2,3. The results show the non-blind and blind scenarios. Each result is accompanied by zoom and PSNR(dB). In blind scenarios sigma = [0.5, 3]. SISR performance of different models on Butterfly image after Gaussian blur at sigma = 2. In blind scenarios sigma = [0.5, 3].