After the back-projection with the color set, the filtered image B'[m,n] is analyzed. The first step in this analysis is a sequential labeling algorithm applied to B'[m,n] that assigns each isolated region a unique non-zero label. Each uniquely labeled region is analyzed in regard to several thresholds. The motivation for these thresholds is to extract regions that meet minimum size constraints and provide a satisfactory representation of the colors in the color set 
that produced the B'[m,n]. If a region does not meet all the thresholds then it is assigned the zero label. The thresholds are described next and are summarized in Table i. These threshold values are appropriate for the full size color images and are rescaled to correspond to the subsampled image size.
The threshold 
corresponds to region size. The region must contain more than = 
pixels to be significant. This value allows for sufficiently small regions to be extracted. The second threshold 
corresponds to the region's absolute representation of each color in the color set. If a selected color from the color set does not contribute at least 
pixels to the region then it is not added. The third threshold 
corresponds to the region's relative representation of each selected color in the color set. It requires that each selected color from the color set contributes at least 
of the region area.
If an extracted color region does not pass one of the above thresholds then it will not be indexed by that color set. If a region is rejected because one of the selected colors from the color set is not sufficiently represented in the region, then the region may still be extracted using a reduced color set that omits the under-represented color. Enforcing the color set thresholds prevents the unnecessary and redundant extraction of multiple-color regions.
binary color set, L[m] = local region histogram.
