Using the circumference will not only be off, but the amount off will vary greatly by how irregular the shape is in X-section in addition to the general curvature.
What I don't like about graph paper is that the usual method (haven't found anything else) is as follows:
1. all lines inside a grid are included
2. all lines outside the grid are excluded
3. all lines through a grid are included at 50%. This is based on the average of 99% in and 1% in, duh.
My problem is that unless you have a 40 foot photo, the grid lines will have far too many lines running through them, and 50% isn't close enough. To get good resolution, you need something the size of a dining room table, like 1/3 scale.
My method:
1. good front photo, high-res
2. re-size it to a multiple of the actual size (width across headlights is easy), 1" = 10 pixels, blah
3. convert to a .bmp
4. fill in the entire outline with a solid color
5. graphics program gives the area of the colored mass in same units as your original measurements
This also allows you to predict area from an actual photo of the real car (not your car necessarily), then lowered, then sectioned, then chopped 6", etc. without doing the work.