-
Notifications
You must be signed in to change notification settings - Fork 1
/
report
30 lines (22 loc) · 1.44 KB
/
report
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
= Computer Vision Project report =
Group (DJ Ninja) members
Aditya Keswani
Ritwik Roy
Algorithm:
Create a descriptor matcher
While the array of images has more than one image:
Compare the first image image with all the other images for keypoint matches, selecting the one with which it has the most matches.
With the first image and the best matched image from the previous step:
Create a blank resultant image and copy the second image to the centre of that result
SURF is used to extract keypoints from both images into two lists, keypoints1 and keypoints2
Keypoints between the two are matched, using a nearest neighbour algorithm to determine.
As a side step, an image is drawn with the two images side by side and lines connecting the matched keypoints
For each matched keypoint pair
Put the keypoint from the first image into a list
Translate the second keypoint to where it should be in the image
Put the translated keypoint into a second list, at the same index/position as its corresponding keypoint in the other list
A homography matrix is generated from the two lists above, using the RANSAC algorithm
This homography matrix is then used to do a perspective warp of the first image into the result (which already has the second image)
The resultant stitched image is cropped and returned
The returned image is put back into the array in place of the first image (and the second image is removed from the array)
Testing procedures