Building the code: There is only a single mex/cpp file included in this package that can be build with the MATLAB function "mex". The steps are: mex -setup C++
mex SR_scale_Hadamard.cpp ("scale" is the upscaling factor,e.g. mex SR_2_Hadamard.cpp ). We provide a pre-compiled file "SR_2_Hadamard.mexw64" for Windows (64 bit).

Using the code: The file "DC_demo.m" is the demo usage of the proposed algorithm. However the file contains one test image (MRI) c03_1.bmp. The remaining test images can be download from http://splab.cz/en/download/databaze/ultrasound (Ultrasound dataset) https://data.mendeley.com/datasets/p9bpx9ctcv/2 (Angiography dataset) https://www.med.harvard.edu/aanlib/home.html ( CT & MRI dataset) https://dl.acm.org/doi/10.1145/3083187.3083212 (Endoscopic dataset) https://stanfordmlgroup.github.io/competitions/mura/ (X-ray dataset)

The training data can be download from above dataset links. The traing data should be kept in a folder "Source1" in the directory. The training dataset used in our paper can be generated by file "create_train_data.m".

The modcrop function is from the source code of: C. Dong, C. C. Loy, K. He, and X. Tang. Learning a deep convolutional network for image super-resolution. In European Conference on Computer Vision, pages 184¨C199. Springer, 2014.

The learned mapping models as well as the decision tree for scaling factor 2 can be directly used for SR reconstruction.

For demo purpose, we have presented results for scale factor 2 in given code. The results for scale factor 4 can be obtained by going through above given instructions. The quantitative results are compared using psnr & ssim for this study.