In this assignment we implement a tool for Image-Stiching, which is useful for making panorama. In this report, we demonstrate the first phase, including feature detection and feature description. Some simple matching results are shown for demonstration.

Team Member
R94725022 資管所 黃輔中

Platform
MS Visual C++ 6.0

Referred Paper
David Lowe, "Distinctive Image Features from Scale-Invariant Keypoints", IJCV 2004.

Algorithm
In this assignment we have chosen SIFT as the feature detection module. SIFT use difference-of-gaussian as the approximate to the Laplacian. The only difference depends on a constant K, which specifies the number of images within a octave. As K increases, the approximation error decreases. The choice of using DOG function is to represent the different scale of the image, as features have the APATURE problem. Further discussion can be found in David Lowe's paper.


Results
The results of the image stiching result are shown right.

First result is the Einstein with its rotated image.

Photo taken at Lab 204.

Photo of a poster.

Taipei 101, photos taken at New Year