Surface modeling methods: correlation vs. unique feature extraction

Scoliosis is a complex deformity of the trunk that induces curvature of the spine and axial rotation of individual vertebrae. Valid 3D models of the trunk surface would enable physicians to evaluate natural history and the effects of treatment for scoliosis. The authors propose a stereo vision system which works as follows. First a slide consisting of light and dark lines is projected onto the back. This slide is coded with white, black, and grey lines. The code was chosen to make correlation results easiest to use. The next step is for the operator to verify that the subject is positioned correctly. This means that the slide is covering as much of their back as possible and that the horizontal reference line is near the middle of their back. Two images are captured of the back at 640 by 480 resolution in 8 bit greyscale. The cameras that capture these images are separated by roughly 40 degrees. This is sufficient for the model accuracy that we desire. After the images are transferred to the main computer, the operator has to specify which section of the image is to be analyzed. The raw images are then analyzed by the model building system.