Description
In this project each student must segment a range image based upon surface normals. A
range image of a chair is given at the course website (note that the reflectance image is
only for visualization and will not be used for the lab; make sure you work with the range
image). Some C-code is also provided to convert the pixels into 3D coordinates. The
segmentation process will use the image grid for grouping pixels, but will use the 3D
coordinates for calculating surface normals for region predicates.
The image should first be masked by thresholding at a distance that removes the
background and leaves only the floor and the chair. Specify the threshold chosen in your
report.
Use the provided C-code or an equivalent translation from the dissertation chapter to
calculate 3D coordinates for the pixels. The slant type can be assumed to be scandirection downward.
Surface normals should be calculated using the cross product method as discussed in
class. Specify the distance chosen between pixels for cross products in your report.
Region growing should be used to segment regions, using the queue-based C code
previously provided. The region predicate should be that a pixel can join the region if its
orientation is within a threshold of the average orientation of pixels already in the region.
The angular difference should be calculated using the dot product. The region growing
code must be modified to recalculate the average after every new pixel joins the region.
Specify the angle threshold chosen in your report. You may take advantage of the fact
that for this image, the surfaces all have strong orientation differences.
Seed pixels for region growing should be found by identifying a complete 5×5 window of
unlabeled (and not masked out in the first step) of still-unlabeled region. If any pixel
within the 5×5 window is masked out or already labeled in a region, then the pixel cannot
seed a new region. Region growing ends when there are no more possible seed pixels.
You must write a brief report that includes the segmented image. The image should be
clearly labeled using either color or visibly different greys. Provide a table specifying the
following for each region: #pixels and average surface normal (x,y,z). Identify the
distance(s) chosen for the cross product for calculating surface normals and the
threshold(s) chosen for removing the background from the image prior to region growing.
Identify the predicate formula used during region growing.
Submit your C-code (as an attachment) and report (as an attachment) to
ece_assign@clemson.edu. Use as subject header ECE4310-1,#8 or ECE6310-1,#8. This
email is due by midnight of the due date.
