Matlab image processing

Non-rigid registration of images

Take your mugshot (template), and take another with a HUGE smile (target)...crop these images

to be the same dimensions. Place corresponding landmarks on each image.

1) Using these landmarks, generate a non-rigid transformation that will transform your smile-free

image towards the one with the smile, and vica versa.

Note 1: Choose a reasonable standard deviation for the Gaussian smoothing. If you choose

something too small, it will create a very local movement of pixels. If you choose something too

large, it will cause lots of pixels even from far away to move with the landmark. The proper

sigma is dependent on the size of your image, and you can find one based on experimentation.

Note 2: Plot the vector field u(x), the transformation , and the inverse

.

2) Place landmarks in the template so that the non-rigid transformation created from these will

generate a motion that is physically not possible - ie say moving your eyes or nose in ways that

humans can't do. Assign the destination of each landmark in a way that you will anticipate will

generate this motion. Transform your image in this way and show before and after

transformation.

3) Create an 'animation' of the overall motion by taking small steps along the displacement and

creating a transformation at each of these steps, apply to the image and save as an animated gif.

Add some fun experiments that try different transformations of your mugshot image...maybe

take a group photo and morph the faces in fun ways....Display an animated gif of this work.