Image Carving

• Algorithm
  • Compute energies
  • Find lowest-energy seam
  • Remove seam
• That's about it. I didn't do image resizing.

Image Stitching

Image Stitching

• Results

Mosaic Source Images

Corners & Matching

Strict feature matching thresholds

Corners & Matching

Looser feature matching thresholds

Stitching 2 Images

3 Images: No Centering

Not the nicest transform ever.

3 Images: With Centering

Easier on the eye.

4 Images

(Low-res because my target image size computation is wrong.)

Image Stitching

• Interesting bits
• How-2

Blending Images

• While blending, maintain a map of how many
  source images contribute to each pixel.
  • Check for black pixels in transformed images.
• Add all transformed images together.
• Divide by count, to average per-pixel.

Optional Parameters

• Parameter tuning is annoying
  • But more annoying when you have to change every parameter in two places (one for debug view, one for real)
• Added optional parameters

function [Y, X, rmax] = anms(cimg, max_pts, ANMS_THRESH)
%ANMS Adaptive Non-Maximal Suppression

if nargin < 3
    ANMS_THRESH = 1000;
end

• Randomly get one perfectly stitched pair of images for no apparent reason, making you think everything is fine except for thresholds.
• Index into arrays of indices using indices found in the array of indices, causing all of your features to match things that don't look anything alike.

How-2: do things wrong

• Forget all 50 warnings about x/y vs. row/col. Wonder for days why some indices need to be flipped, but keep going anyway.
• Pretend to understand the difference between indexing with integer matrices and indexing with logical matrices, but don't.
• Have MATLAB bring your computer to a screeching halt by using 16 GB of memory on a system with 8 GB of RAM.

How-2: do things wrong

bloopr

"bloopr: it's like bad Instagram filters for MATLAB"