The next stage involves the extraction of the color regions from the images and is illustrated in Figure 8. This is done by systematically iterating over color sets for the image. The back-projection of the binary color set onto the image produces a bi-level image B[m,n]. The levels correspond to the selected and un-selected pixels for the specified color set.
After back-projection the bi-level image is filtered morphologically to (1) eliminate spots and (2) connect neighboring regions. The binary color set back-projection algorithm is given as follows: given binary color set 
and processed image R[m,n]:
: let k = R[m,n] then B[m,n] = c[k].
box median filter U[m, n] to produce B'[m,n]. Since B[m,n] is a bi-level image, the median statistic is equivalent to the majority computation.
Figure 8: Second and third stages of color region extraction: (1) back-projection of color set (in this example has one color selected) onto processed image R[m,n] produces bi-level image B[m,n], (2) filtering with U[m,n] produces B'[m,n] and (3) thresholding determines surviving regions.