This project implemented this paper,
Wang, M., Wang, B., Fei, Y., Qian, K., Wang, W., Chen, J., & Yong, J. (2014). Towards Photo Watercolorization with Artistic Verisimilitude. IEEE Transactions on Visualization and Computer Graphics, 20, 1451-1460.
The results mentioned in this paper was not fully reproduced in this project. But we have achieved similar effects and obtained desirable results for some images.
This project was originally written in Arch Linux, with CMake version 3.13.2, GCC version 8.2.1, and OpenCV version 4.0.1.
- GCC (with C++17 support)
- OpenCV
- CMake
To build this project, cd into the build directory and type:
cmake ..
make
To build the clustering tool, cd into the ClusterTool directory and type:
mkdir build
cd build
cmake ..
make
To apply watercolorization effect to an image:
./Watercolorization /path/to/inputfile [/path/to/outputfile]
If /path/to/outputfile was not given, the watercolorized image would be output as output.jpg in the working directory.
To use the clustering tool, place the training images into PicutreDatabase/Original and run:
./Clustering ../../PictureDatabase
The pictures will be classified into 20 classes. Folders corresponding to each class will be created, with the symbolic links to the original files automatically generated in them. The trained centers of the k-means algorithm will be stored in model, which should be put in the same directory of Watercolorization in order to enable the auto-detection of the closest style of images.
This project was licensed under GNU GPL3. Check the LICENSE file for more details.