Pixel-art Upscaling

This repository contains C++ and Python implementations of various algorithms for upscaling pixel-art as part of the final project for CS4365 Applied Image Processing. For further details, kindly refer to the provided report.

Group Member(s)

• William Narchi (5046122)

Requirements

• Python 3.10 or higher (the implementation has been tested on Python 3.10.6 only, though is likely to work on older versions)
• CMake 3.11 or higher
• C++20 compliant compiler
• OpenMP 2.0 or higher
• Cairo graphics library

Build Instructions

The C++ portion of the codebase uses CMake, so simply use your favourite CLI/IDE/code editor and compile the fin-proj target. All dependencies are included in this repository and compiled as needed.

For the Python portion of the codebase, simply install the packages specified in requirements.txt and run main.py from the root of this repository. You must ensure that the Cairo graphics library is installed as well.

Directory Structure

• framework contains a slightly modified version of the framework used by the Computer Graphics and Visualisation group at TU Delft for the assignments for CS4365 in addition to the following external libraries
  • catch2
  • Eigen
  • fmt
  • glm
  • stb
• src contains concrete implementations
  • C++
    • 2xsai.hpp contains an implementation of the '2x Scale and Interpolate Engine' by Derek Liauw Kie Fa
    • common.hpp contains functionality used across several of the implemented algorithms
    • eagle.hpp contains an implementation of the Eagle upscaling algorithm
    • epx.hpp contains an implementation of the 'Eric's Pixel Expansion (EPX)' upscaling algorithm by Eric Johnston and the 'AdvMAME2x' algorithm
    • hq2x.hpp contains an implementation of the hq2x upscaling algorithm by Maxim Stepin
    • nedi.hpp contains an implementation of the 'Adaptive New Edge-Directed Interpolation' algorithm by Fan-Yin Tzeng, which is based on the 'New Edge-Directed Interpolation' algorithm by Xin Li and Michael T. Orchard
    • xbr.hpp contains an implementation of the 2x version of the xBR algorithm by Hylian
  • Python - implementation of the Kopf-Lichinski pixel-art upscaling algorithm
    • geometry.py contains functionality for creating and manipulating B-spline curves
    • heuristics.py allows for the computation of the diagonal edge cost heuristics specified in section 3.2 of the paper
    • kopf_lichinksi.py contains the bulk of the implementation, primarily localised within the Vectorizer class
    • main.py is the main entry point to the application and contains the code used to generate the data in the report
    • pixel_io.py contains functionality for loading images for upscaling (24-bit PNGs without an alpha channel) and outputing both SVG files and rasterized PNG files
    • utils.py contains various utilities used throughout different parts of the implementation

Credits

The Python implementation is heavily based on

• This Python implementation of the Kopf-Lischinski approach by Campbell Barton
• This Python 3 adaptation of Campbell Barton's implementation by Anirudh Vemula and Vamsidhar Yeddu