For automatic processing of mail, the problem of locating the destination a
ddress block (DAB) on the images of mail pieces is still a bottleneck Using
a bottom-up approach, we locate the DAB in allusion to complex mail pieces
. Our algorithm starts from local binarization between the rising and faili
ng points in the image, following by searching, filtering, feature extracti
on, and modified region growing of connected components. Finally, some feat
ures of the candidate blocks are extracted for assigning weights to locate
the DAB. The success of our algorithm for locating the DAB on complex mail
pieces is demonstrated on 100 test images with the success rate 90.7%. The
results also show that our approach can process a DAB which is skewed, hand
-written, or machine-printed. The average processing time for each piece im
age of size 720x960 is 2.66 seconds on a PC (PentiumII-400).