Jigsaw Puzzle-Solving Computer Makes Us Feel Inadequate
Taeg Sang Cho and his colleagues at the MIT aren't the first to write jigsaw puzzle-solving software; a Danish team wrote a program that was able to solve a 320-piece puzzle back in 2008. While the older program only worked on simple "cartoon-style" drawings with a limited color palette, Cho's program can solve puzzles created from complex images (like landscape photographs), and handle an almost unlimited number of colors. Cho's creation recently solved a world record-setting puzzle with 400 pieces, beating the previous effort for computerized jigsaw solving (yes, apparently there are records for such an achievement). Previous software used standard jigsaw shapes that give additional clues to placement. Cho's method (.PDF) used simple squares, forcing the AI to rely entirely on color and pattern recognition.
The software first scans for broad color patterns, and by comparing those to a database of known images, arranges the pieces in an approximation of the final picture. It then analyzes the color of pixels around each piece's edge to find other pieces that might match and starts assembling the picture. In less than three minutes, the jumbled pile of squares is a flawlessly reconstructed photo.
Cho believes his technology will eventually find its way into photo-editing software, where the new color matching and edge detection techniques could create more realistic manipulations. It could also alert users to inconsistencies caused by shoddy editing, taking all the fun out of Photoshop Disasters. [From: Popular Science and New Scientist]