Implement a solver based on edge curvature of pieces

[chrismwendt]

• construct a list of edge curvatures for each piece
  • run blob detection on the piece with a threshold of 50%
  • construct a list of edge points by walking the perimeter
  • construct a list of edge curvatures by computing the curvature at each edge point as the distance to the line of best fit of nearby edge points
• pick an arbitrary starting piece and call it the "island"
• attach the rest of the pieces to the island one by one, always attaching the best match next
  • the best match has the smallest minimum error of its shifted positions against the island - a shifted position is an alignment of a piece along the island's edge
  • the error of a shifted position is the minimum of (sum of contiguous differences between its curvature and the island's curvature, divided by the length of the subsequence)

[chrismwendt]

[oggnik]

Finding minimum error of piece A compared to piece B:

if (A.length < B.length)
  switch A,B
  remember to return the right things

for (shift = 0; shift < A.length; shift++)
  for (i = 0; i < B.length; i++)
    add B[i] + A[i+shift] into new list
  find max sub array to find section with smallest error
      this returns the indexB, length, error

indexA = shift+indexB

keep track of index's, length, and error for smallest error

return:
indexA
indexB
    switch if the lists were switched
length
error

return as a class defined inside of utilities

[chrismwendt]

Sections of the perimeters of pieces are able to be matched. See 42ca3066100aec387a2b448e2d46a45e17705ded. Still need to iteratively attach pieces to the island.

[oggnik]

I added the iteratively adding pieces to the island. My best matching so far:

[aarya123]

omg, this is so intense!