Assumption: transparent traceable “human figure” in 2d image
The way to do this is to require the user to take two photos. One image is an image with the image of the user, another image must be executed in the same camera position and orientation, but the user leaves the frame for this.
Since both images will have the same background, you can compare pixel by pixel between two images and mark those pixels that have a difference above a certain threshold. Of course, the threshold should be chosen so that camera noise is not detected as a difference. When you have a collection of pixels that are different from each other, you can filter them out and calculate the approximate silhouette for the user from pixels around the edge.
Simplification of the above method can be performed if you have control over the background. You can use bluescreen to avoid having a second image with a background.
Miguel
source share