Visual defectives inspection and classification using a multi-layer perceptron

This paper proposes an efficient feature representation scheme for a visual inspection problem which uses curvature and an iterative circular fitting algorithm. Homogeneous ranges as well as critical points on the boundary such as vertices, inflection points, linear, and circular features can be detected. Thus, a shape boundary can be represented by the combination of linear and circular features. For the classification purposes, a multi-layer perception which takes separated local features as input parameters, has been adopted. The test results showed that they are robust within some error range.