This repository contains a Python implementation of the algorithm described in the paper Least-Squares Image Resizing Using Finite Differences. The algorithm provides an optimal spline-based approach for resizing digital images with arbitrary (non-integer) scaling factors. It minimizes artifacts such as aliasing and blocking, improving the signal-to-noise ratio compared to standard interpolation methods.
- Optimal spline-based image resizing
- Arbitrary scaling factors
- Consistent reduction of artifacts and improved signal-to-noise ratio
- Computational complexity per pixel is independent of the scaling factor
Resize/Resize.py: Core library implementing the resizing algorithm.Original_Source_Java_Code/: Directory containing the original Java implementation of the algorithm.Samples/: Directory containing sample images for testing.Results/: Directory containing plots and result images from the comparison tests.testComparison_LS_Arrate_Ground_Truth.py: Comparing the code results with the original Java source.testComparison_LS_Interp_Oblique_Plot.py: Compares LS, basic interpolation and Oblique interpolation across several scaling factors.testComparison_LS_Interp_Oblique.py: Compares LS, basic interpolation and Oblique interpolation.testComparison_LS_Linear_Cubic_Plot.py: Compares LS between Linear and Cubic interpolation.testComparison_LS_Oblique_Scipy_Plot.py: Compares Oblique interpolation and Scipy interpolation across several scaling factors.testComparison_LS_Scipy_Plot.py: Compares LS and Scipy interpolation across several scaling factors.testComparison_LS_Scipy.py: Compares LS and Scipy interpolation.
To use the core library, only python and numpy is required. To run the test examples, you will need additional dependencies.
-
Clone the repository:
git clone https://github.com/splineops/splineops-experimental.git cd splineops-experimental/Resize -
Install the required dependencies:
pip install numpy scipy matplotlib imageio
-
To run the tests, just execute them with Python.