Professional Master Program - UW
Computer Vision - CSEP 576
Project 3: Photometric Stereo
Student: Marcelo De Barros

Buddha

RGB

Needles

Albedo Map Snapshot

View 1 without Albedo

View 2 without Albedo

 

View 1 with Albedo

View 2 with Albedo

Horse

RGB

Needles

Albedo Map Snapshot

View 1 without Albedo

View 2 without Albedo

 

View 1 with Albedo

View 2 with Albedo

 

Owl

RGB

Needles

Albedo Map Snapshot

View 1 without Albedo

View 2 without Albedo

 

View 1 with Albedo

View 2 with Albedo

 

Marcelo!

Well, this was a tentative of doing the photometric stereo of my own picture, but the results don't look quite well as I would like. Some reasons could be:
 - The intensity of the light source was not uniform over the object
 - The direction of the light source varied slightly between the time I took the pictures with the sphere and when I took the pictures of my face
 - It seems that the intensity of the light on the right side was greater than the one on the left side
 - I moved slightly during the pictures, and so did the camera
I took 12 pictures of myself and I used one of those xmas ornament balls as the sphere to calculate the light directions.
Then using the intelligent scissors tool, I created the mask for both the sphere and my face. Then I just tried using the tool against these images.
I figure since I spent a good one hour taking all the pictures, I might as well just add the bizarre results to the artifact page...

Marcelo - original picture (first image out of 12)

Sphere to calculate the light directions (first image out of 12)

RGB

Needles

Albedo Map Snapshot

View 1 without Albedo

View 2 without Albedo

 

View 1 with Albedo

View 2 with Albedo

About the project!

I had a lot of fun, it was a nice project! In the beginning it was tough to understand the concept, but once you get the idea the implementation is very straightforward. I liked the fact that we only had to fill out the gaps related to the subject and didn't have to spend too much time in coding the math routines (although it would be definitely interesting too!). The depth map was by far the most complicated part, at least for me. It took me a while to realize that the zind vector already contained the indexes to be used in the M matrix. I also had to be careful to stay within the bounds of the mask and to avoid NaN. After that, the project was finally done!
The 3D for most of the images turned out being very good. The image of the owl did not work very well for the right side, perhaps because there was not much light on that side of the pictures. Of course, my image was not reconstructed properly either due to the reasons already mentioned. Two problems with this method are: first, it gives extra weight to the areas with more illumination, and second it leaves a trail of the image when moving it around after the depth map has been calculated. Unfortunately I'm not sure how to fix these two problems.