juefeix@andrew.cmu.edu
Oct.1 2010
I. Toy Problem
In this problem, both x direction and y direction gradients from an image s is computed. I then use the gradients plus one pixel intensity to reconstruct the image v.
The methodology is to solve a set of least squares constraints. In this problem, 3 objective functions are needed to solve the constraints.
Here is the original image and constructed image.
Original image:
Constructed image:
Constructed error: 0+4.7186e-006i
II. Poison Blending
Just like problem 1, poison blending can be implemented by solving least square constraint. This time, I use 5 objective functions. 2 for gradients at x direction and 2 for gradients at y direction and plus 1 pixel intensity function. Since I am dealing with color image this time, I am treating each channel separately.
Here are the example of my experiment.
(1) Eye on hand
(2) Penguin on snow 1
(3) Penguin on snow 2
(4) Pisa tower and Hamerschlag Hall
The last blending is a bad one because the pixel intensity between the source and background are way different from each other.
III. Mixed Gradient Blending
Similar to Poison Blending, Mixed Gradient Blending uses the gradient in source or target with the larger magnitude as the guide, rather than the source gradient.
Here are 2 examples for mixed gradient blending:
(1) Penguin on snow
(2) Google logo in night sky
IV. Color2Gray
I tried to convert the color image into HSV channel, and use Mixed Gradient Blending to blend the 2nd and 3rd channel. Because 2nd channel in HSV is satuation and 3rd channel is value. So 3rd channel should be source and 2nd channel should be background. Here is the result.
END