This repository is related to a MATH 214 class project at the University of Michigan: Linear Algebra and Least Squares Matching
We have implemented LSM image matching in Python using image and matrix manipulation libraries including OpenCV and NumPy.
The algorithm above is used in the program and takes for reference a lecture taught by Professor Cyrill Stachniss at the University of Bonn, Germany in the summer term 2015 <Lecture Page 30>. For simplicity, we assume there is no noise in images. i.e. The v term is the zero vector.
Make sure the original image is named "OLD.png", the distorted image "NEW.png", and run with
python main.py
Close opened images as it runs. The result is saved as "DONT_TOUCH_ME.png"
As is shown in Fig. 1, the distorted image (mid-left) was 10 pixels left and 15 pixels up compared to the original image (top-left). Lost areas are filled with edge pixels. After running 10 iterations of our LSM process on the distorted image, we obtained the corrected output (bottom-left). Comparing the original-distorted overlay (top-right) and the original-corrected overlay (bottom-right), we observed that LSM is effective in aligning similar images.
