Automatically Generated Morph Sequence

A computer vision based morph between 6 hot celebrities, and 6 random women from Google Image Archives, (perhaps on their way to becoming celebrities themselves). Our vision system optimized the sequence of faces and the smoothness of each transition pair in order to morph without the usually required manual-entry of control point pairs. The sequence optimizations were implemented using Simulated Annealing to improve a Greedy Tour as an approximation to the NP-hard Travelling Salesman Problem, which must find an optimal Hamiltonian Circuit through the graph of face-nodes among the exponential possibilities of such tours. You can tell that even our approximated system is wise, because Lindsay Lohan is morphed into Dar Williams, producing a few frames of what could only be otherwise achieved through decades of expensive cloning and DNA splicing procedures. This is a Final Exam Team Project for the University of Pennsylvania's GRASP Robotics Lab course: CIS-581, Computer Vision. Taught by Professor Jianbo Shi, Fall 2008.




Authors: Srinivasan Dwarakanathan, Hua-Min Shen, Jonathan Schmid.

Category:

Science & Technology

Tags:

• Automatically
• Generated
• Morph
• Sequence
• Female
• Celebrities
• Simulated
• Annealing
• Hamiltonian
• Circuit
• Approximation