Presidential Face Morphing

Assignment for cs195-g, Computational Photography at the Brown University Computer Science Department, Spring 2010.

The results were produced automatically using an image morphing algorithm implemented in MATLAB. The algorithm requires correspondence points selected by a user (in this case, me). The results would have looked been better had I aligned the faces or been more consistent choosing images. However, the poor choice of portraits shows many of the limitations of the algorithm.

The algorithm cross dissolves the colors using a simple alpha blend method. The algorithm smoothly warps the two images by performing a Poisson 'image' fill using the offsets of each image's correspondence points to current linearly interpolated points as initial values to get the pixel indexes into the rest of each image.

This video was produced over 2 days: writing the algorithm, selecting the images and 22 correspondence points per image, and running the algorithm to produce 60 frames per morph on a quad-core machine (2+ hour runtime).

Category:

Education

Tags:

• President
• Morph
• Brown University
• Computational Photography

[markmh835]

It really is more fade-out, fade-in, than true morphing. 

[JanetandAutumn]

most of the images are facing a certain way except lincolns