README.md

Image Warping with TPS and CSRBF

This repository provides a simple framework for flexible image registration and warping using two common methods:

• Thin Plate Splines (TPS)
• Compactly Supported Radial Basis Functions (CSRBF)

The design follows an object-oriented pattern, with a shared BaseTransform class and specialized subclasses (TPSTransform, CSRBFTransform).
This makes it easy to extend with additional warping methods.

---

Usage

from skimage.io import imread, imsave
from Transform import TPSTransform, CSRBFTransform

img = imread("boats_grid.jpg")
h, w = img.shape
anchor = [[0,0], [w, 0], [0, h], [w, h]] # anchor at four corner of image

# expand and shrink at two location
src = [[145, 130], [220, 130], [145, 180], [220, 180], [410, 317], [604, 314], [413, 471], [605, 472]]
dst = [[115, 90],  [265, 90],  [115, 245], [270, 245], [445, 350], [570, 350], [445, 445], [570, 445]]
src += anchor
dst += anchor

tps = TPSTransform()
warped = tps.transform(img, src, dst, order=0)
imsave("boats_tps.jpg", warped)

csrbf = CSRBFTransform(200)
warped = csrbf.transform(img, src, dst, order=0)
imsave("boats_csrbf.jpg", warped)

Citations

1. Skimage ThinPlateSplineTransform
2. Principal warps: Thin-plate splines and the decomposition of deformations
3. Radial basis functions with compact support for elastic registration of medical images