In a computer-implemented method for detecting three-dimensional surface geometries of real intraoral structures, two-dimensional images of intraoral structures are converted into three-dimensional virtual surface geometries. Pixels of the two-dimensional images are each partitioned into segments and each segment is assigned a label. Each segment of a two-dimensional image corresponds to at least one type of real intraoral structure. There are at least two types of segment; each type of segment has a different label and at least one of the labels represents teeth and tooth-like structures.