Given a set of corresponding features, we use the planar Homography to register multiple images. The proposed parameter estimation algorithm outperforms conventional method for the Homography estimation problem.
Example application: retinal image registration and 3D surface reconstruction [Link]
The paper proposes a new image feature extraction algorithm, Locally Salient ICA (LS-ICA), that represents locally salient information from important parts of images. The algorithm was tested on facial images in the context of face recognition.
The proposed optimization framework uses a new photometric consistency metric among multiple images to produce the geometry of the object (face).
The proposed image transform method (from RGB image to another color space) provides color-based segmentation results.
Note: This is not my work but I partly supported this research focusing on the image registration.