Leather inspection through singularities detection using wavelet transforms

Abstract

The major problem in leather inspection is to separate defects from the background exhibiting a wide range of visual appearances. Leather defects, characterized by oriented structures, cannot be easily discriminated from the structures typical of the normal surface. Though gaussian filters generally represent a successfull tool to smooth out the structures on the background, a wrong choice of the resolution can preclude the detection of defective regions (singularities) in the subsequent analysis. However, wavelet transforms can be profitably used for studying the evolution of singularities across different scales. Suitable kernels for this transform does allow multiscale singularities analysis through the detection of local maxima in wavelet transform modulus.