Project 1 Writeup

Alex Collins (aecollin)
January 29, 2010


The overall algorithm creates an image pyramid by recursively scaling the image until its size is less than 256x256 pixels. At this point, the image is small enough that we can exhaustively search through many different shift values to determine the shift that minimizes Sum of Squared Difference (SSD) of the intensities of each color plate. Parameters:

Discussion of Results

Nearly all of the images lined up extremely well using the brute-force search, which was to be expected. When using the Image Pyramid method, the majority of the images still lined up very well, however many have slight color blurring due to imperfect alignment. A couple images, such as 00153v.jpg, never managed to have all 3 plates line up correctly. This may be because these scenes are lower contrast, and so there may be much less differentiation in the intensities of the color plates.

One interesting discovery was the the .jpg images aligned best when I reversed the xy red-shift values. However this was not a bug in the program since the .tif images aligned best using the calculated red-shift xy values.

Extra Credit

Result Images

JPG Images aligned using Brute Force (Left) and Image Pyramids (Right)

TIF images aligned using Image Pyramid Methods

Red-Shift (y,x) and Green-Shift (y,x) data for all images Image Name
Green Shift (brute force) Red Shift (brute force) Green Shift (pyramids) Red Shift (pyramids)
00125v.jpg (5,2) (10,1) (4,2) (4,8)
00149v.jpg (4,2) (9,2) (4,2) (11,4)
00153v.jpg (7,3) (-41,36) (4,4) (-26,30)
00154v.jpg (5,0) (11,-2) (2,2) (8,0)
00163v.jpg (-3,1) (-4,1) (0,0) (-2,0)
00270v.jpg (3,1) (12,1) (2,2) (10,0)
00398v.jpg (5,3) (11,4) (4,4) (14,6)
00458u.tif N/A N/A (38,10) (86,42)
00564v.jpg N/A N/A (2,4) (10,2)
00911u.tif N/A N/A (6,10) (118, -6)
001043u.tif N/A N/A (-6,6) (14,14)
001057u.tif N/A N/A (22,22) (54,42)
01167v.jpg N/A N/A (2,4) (10,0)
01657u.tif N/A N/A (38,26)) (102,26)
01861a.tif N/A N/A (54,42) (134, 74)
31421v.jpg N/A N/A (2,6) (2,15)