TUTCRIS - Tampereen teknillinen yliopisto

TUTCRIS

Shape and gray level descriptors for surface defect image retrieval and classification

Tutkimustuotos

Yksityiskohdat

AlkuperäiskieliEnglanti
JulkaisupaikkaTampere
KustantajaTampere University of Technology
Sivumäärä62
ISBN (elektroninen)952-15-1497-3
ISBN (painettu)952-15-1462-0
TilaJulkaistu - 25 marraskuuta 2005
OKM-julkaisutyyppiG5 Artikkeliväitöskirja

Julkaisusarja

NimiTampere University of Technology. Publication
KustantajaTampere University of Technology
Vuosikerta556
ISSN (painettu)1459-2045

Tiivistelmä

Nowadays, the problem of image retrieval and classification plays an important role in the fields of image analysis and pattern recognition. With an increasing amount of real-world image data to be processed and stored, the development of powerful retrieval tools has also become a central problem in various industrial machine vision applications. The goal of finding similar objects from large and often distributed image collections is shared by the developers and users of machine vision systems. The focus of this thesis is on the field of surface defect imaging that has been applied to paper and metal manufacturing. Current surface inspection systems are capable of detecting various defects and producing gray level images of them. The defect images are collected into large image databases. Effective retrieval and classification methods are necessary to analyze the defects stored in the database. The goal of this thesis is to present visual descriptors that characterize the defect shape and gray level distribution. The majority of the methods presented consider the shape using Fourier description of the boundary line of the defect. For this kind of shape description, novel Fourier-based approaches are presented. These approaches add a multiresolution property to conventional Fourier shape descriptors. This is achieved by combining discrete wavelet transform with discrete Fourier transform in shape description. Another approach to multiresolution shape description uses boundary smoothing combined with Fourier shape description. In addition, a method to combine defect boundary with defect s gray level information in Fourier description is presented. The gray level distribution of a defect image is described using binary co-occurrence matrix that outperforms commonly used second order statistical measures in defect image retrieval. The proposed shape descriptors provide a significant improvement over the conventional Fourier shape description of the defects. The experimental results reveal that retrieval accuracy can be easily improved by using the proposed multiscale Fourier descriptors. The descriptors which use a combination of defect shape and gray level provide a novel method that is capable of improving retrieval performance without increasing descriptor dimensionality.

Julkaisufoorumi-taso

Latausten tilastot

Ei tietoja saatavilla