We are concerned with extracting motion information from difference im
ages. A difference image, formed by comparing gray-levels of two conti
guous frames, contains moving areas. These moving areas are due to eit
her COVering or UNCOVering a background by a moving object. We first d
escribe how to label a moving area either COV or UNCOV. Especially, it
will show how to subdivide a moving area formed by multiple moving ob
jects, into COVs and UNCOVs of each object. We next extract motion cha
racteristics of each COV and UNCOV region by examining distributions o
f intensities and topological structures. Experimental results show th
at the suggested method is very robust and accurate for natural as wel
l as lab-generated images.