This folder contians the executable codes for the paper "Learning Hierarchical Decision Trees for Single Image Super-Resolution".

Module Leader: Prof. Wan-Chi Siu, PhD, CEng, FIEE, FHKIE, FIEEE, Chair Professor
M.Phil/PhD Student: Mr. Jun-Jie Huang, BEng

Version 1.0
- First release
January 2016

Version 1.1
- Numerical results on Set5, Set14, and BSD100 added in Results.xlsx
May 2017


==============================================================
About this folder
==============================================================
SRHDT.exe is the executable code for our SR algorithm.

The TesingImages.txt indicates the number of testing images and the file names of testing images(the bicubic interpolated image using MATLAB function imresize). For example, the "14" on first line means that there are 14 testing images. And the "baboon[2-Bicubic].bmp" on the second line means the first testing image is baboon[2-Bicubic].bmp.

The decision trees in the folder "Models\2xTrees-118" and "Models\3xTrees-118" are trained using our selected 118 training images for upscaling factor of 2 and 3, repectively. The decision trees in the folder "Models\2xTrees-91" and "Models\3xTrees-91" are trained using the standard 91 training set for upscaling factor of 2 and 3, repectively. Each folder contains 4 decision tree tables.

2xSet5-Bicubic contains the gray level images of the original set5 images and their bicubic 2x downsampled then 2x upsampled images.

3xSet5-Bicubic contains the gray level images of the original set5 images and their bicubic 3x downsampled then 3x upsampled images.

2xSet14-Bicubic contains the gray level images of the original set14 images and their bicubic 2x downsampled then 2x upsampled images.

3xSet14-Bicubic contains the gray level images of the original set14 images and their bicubic 3x downsampled then 3x upsampled images.

2xBSD100-Bicubic contains the gray level images of the original BSD100 images and their bicubic 2x downsampled then 2x upsampled images.

3xBSD100-Bicubic contains the gray level images of the original BSD100 images and their bicubic 3x downsampled then 3x upsampled images.


GenerateBicubicDegradatation.m can generate bicubic downsampled then upsampled images using the MATLAB function imresize.

==============================================================
Notes
==============================================================
The testing images should be the bicubic interpolted images using MATLAB function imresize.

==============================================================
How to test the proposed image super-resolution method
==============================================================
Put the testing images in the same folder as the SRHDT.exe

Then, run SRHDT.exe

The program will first load the needed tree tables. Then the testing images will be loaded and tested one by one.

==============================================================
Results
==============================================================
The printed processing time indicates the computation time for each testing image in ms. 

The n times upscaling result of the Layer k decision tree will be stored in the folder "nxLayerk".
