 |
The goal of this project was to implement the algorithm presented in the paper "Automating Joiners" by Lihi Zelnik-Manor and Pietro Perona. The resulting image is supposed to mimic the images produced by David Hockney. The algorithm used in this project is very similar to the algorithm used in the Panorama Project. Given a folder full of images, the algorithm firstly matches the images that are supposed to overlap and than lays them on a canvas. The challenge in this project was to figure out the order of the in order to conserve the continuity/smoothness of the final image.
|
 |
 |
||
| Image Processing Techniques | ||
| The algorithm consists of 5 parts:
1. Detecting feature points through "Scale-invarient feature transform" (SIFT) algorithm. 2. Extract a feature descriptor for each feature point and match the descriptors between images. 3. Use RANSAC to get rid of the outliers. 4. Compute affine/rigid-body transformations and align the images. 5. Change the order of the images through the continuity of the energy functions.
Below, you can see the difference between the ordered and unordered final results:
|
||
|
