Machine vision and automatic surface inspection has been an active fie
ld of research during the last few years. However, very little researc
h has been contributed to the area of defect detection in textured ima
ges, especially for the case of random textures. In this paper, we pro
pose a novel algorithm that uses colour and texture information to sol
ve the problem. A new colour clustering scheme based on human colour p
erception is developed. The algorithm is training based and produces v
ery promising results on defect detection in random textured images an
d in particular, granite images.