Computer Vision: Image mosiacs project

A presentation on the results of a image mosiac project from Computer Vision class.

Transcript

1. Computer Vision Project 3 Image mosaics Ly, Pant
2. Decisions • ANMS ◦ Not really working until we used imregionalmax. • Descriptors ◦ Rotations using atan Iy./Ix • Ransac ◦ 50 random pulls of 4 points • Mosaic merging ◦ Homography applied to all pixels of source image ◦ Results in holes in destination
3. ANMS - 50
4. ANMS - 200
5. Test Mosaic
6. Bridge Mosaic (Ransac poor)
7. Bridge Mosaic (Better--luck?)
8. Chicago Mosaic (3-images)
9. Quad Mosaic
10. Quad Mosaic (Fixed later)
11. Tips • Skipped ANMS by using other code for non-maximal suppression in order to work on other tasks • Did not implement inverse homography to copy pixels ◦ Results in holes in image, especially when there are perspective transformations (Quad Mosaic) ◦ (Later implemented.) • Perseverance!
12. Problems • ANMS slow. • Crashed matlab by not setting bounds on images ◦ Too much memory consumed • Did not implement inverse homography to copy pixels ◦ Results in holes in image, especially when there are perspective transformations (Quad Mosaic) ◦ (Implemented later using inpolygon) • Ran out of time!