Compcam 2012 Fall Assignment 3

Used the matlab code to do the edge detection. We then do region filling to create mask for foreground/background separation. Kmeans clustering method was applied to create a cartoonized photo, which is used to be the color source of final output image.

Tested on photos given, edges found:

Mask created via region filling:

Cartoonized effect via Kmeans clustering with 8 colors:

Output image (background color changed to illustrate the background/foreground separation):

Another experiment by our own photos:

Since the lighting condition, background is less controlled compared to the photos given, edges detected are more noisy. Also, some edges such as desk boundary is actually correct.

Via low-pass filtering some manually editing (e.g., for desk boundary) we get the following mask:

We again apply Kmeans clustering to get cartoonized effect (with 32 colors):

Here is the output image:

Compcam 2012 Fall Assignment 1

Original photo with left light:

Relighted photo with left light:

Original photo with right light:

Relighted photo with right light:

Original photo with both lights:

Relighted photo with both lights:

Source code can be found in the bitbucket.

Challenges: hard to take photos in position, angles that almost identical without a tripod.

Photo taken with Sony NEX-5N.

About the Title

“Beauty is our business” is a birthday salute to computer scientist Edsger W. Dijkstra. I borrowed the quotation to show that beauty can also be one of the main target of technology.