Potential matches between images lie on epipolar lines. Incorporating a uniqueness constraint, primitives may only take part in one match. Primitives must also be matched in the order they appear in the raster line. For example, in Figure 7, below, primitive L1 in the left image has been matched with primitive R2 in the right image. This now makes it physically impossible to attempt to match the next left primitive L2 with R1, however R3 is a possibility because it appears as the next primitive in order.

To find potential matches requires considering a primitive in one of the images then searching on or either side of the corresponding position in the other image for similar primitives. The area searched in the second image is within the disparity range, which can be calculated normally from the known camera geometry and possible object positions. Whether the left image is considered first and disparity range searched in the right image or vice versa makes no real difference.

Besides lying within the disparity range and gradient a potential match must have similar characteristics. This isn't a problem when matching stereograms since dots are identical and only vary in position within the image. With natural images, it is usual to assume that matched edges must have orientations (and possibly magnitudes) which are sufficiently similar.