Inspection of castings is performed manually, because castings have complic
ated shapes and often include a variety of defects. However, this method of
inspection is often less reliable due to reasons such as operator fatigue
and changes in the environment. A method for automatic inspection of castin
gs using image processing is described in this paper. Here, a method to red
uce noise attributable to surface roughness is first introduced. This techn
ique involves uniform illumination using several ring-shaped fluorescent la
mps. Next, an image processing-based band-pass filter, which is used to dis
tinguish between defect signals and surface roughness-induced noise, is dis
cussed. This filter is composed of averaging and differential operators, an
d improves the signal-to-noise ratio, from 1.67 to 5.35. Finally, a statist
ical method, to distinguish between actual defects and 'pseudo' defects res
ulting from steps, edges, dimples and embossed characters on the casting su
rface, is described. In this method test pieces are compared with statistic
al data obtained from several good castings. When applied on lint, the prop
osed method could be used to successfully distinguish between good and defe
ctive castings.