Image Super-Resolution using Convolutional Neural Network.

SRCNN is a Implementation of Image Super-Resolution using SRCNN which is proposed by Chao Dong in 2014.

• If you want to find the details of SRCNN algorithm, please read the paper:

  Chao Dong, Chen Change Loy, Kaiming He, Xiaoou Tang. Learning a Deep Convolutional Network for Image Super-Resolution, in Proceedings of European Conference on Computer Vision (ECCV), 2014

• If you want to download the training code(caffe) or test code(Matlab) for SRCNN, please open your browse and visit for more details.

• And thank you very much for Chao's work in SRCNN.

This implementation uses a block algorithm. This is due to memory consumption in CNN. For complete processing, memory is required for 175 original images. With block processing, this size is reduced to 170 block size + 5 size of the original image. That allows you to process even large images. At the same time, block processing gives minor block artifacts, which I could only detect using dssim.

submodules:

• stb -> src/stb

$ git submodule init
$ git submodule update

build dependencies:

• build-essential
• cmake

$ sudo apt-get install build-essential cmake

$ mkdir build
$ cd build
$ cmake ..
$ make

The first and second parameters specify the paths to the image and the result {PNG}. The -b option specifies the block size in block division (default 256). The -o option specifies the block overlap (default 16 [~ blok size / 16]). The -p option specifies the part of the CNN in the resulting image (default 0.707107, full 1.0).

./stbsrcnn ${IMAGE_PATH} ${IMAGE_PATH}.out.png

• bicubic.h - bicubic image scaling
• convdata.h - CNN weight data
• dependencies.c - API stb
• srcnn.h - CNN apply
• stb/ - stb
• stbsrcnn.c - CLI program.

Origin (256x256:3):

Bicubic interplation (x2, 512x512:3):

SRCNN (only x2, 512x512:3):

You can run the following command to test the demo program.

stbsrcnn -p 1 ../images/butterfly.png butterfly.x2.png