Generalized Completed Local Binary Patterns for Time-Efficient Steel Surface Defect Classification

Abstract

Abstract-Efficient defect classification is one of the most important preconditions to achieve online quality inspection for hot-rolled strip steels. It is extremely challenging owing to various defect appearances, large intra-class variation, ambiguous inter-class distance, and unstable gray values. In this paper, a generalized completed local binary patterns (GCLBP) framework is proposed. Two variants of improved completed local binary patterns (ICLBP) and improved completed noise-invariant local-structure patterns (ICNLP) under the GCLBP framework are developed for steel surface defect classification. Different from conventional LBP variants, descriptive information hidden in nonuniform patterns is innovatively excavated for better defect representation. This work focuses on the following aspects: First, a lightweight searching algorithm is established for exploiting the dominant nonuniform patterns (DNUPs). Second, a hybrid pattern code mapping mechanism is proposed to encode all the uniform patterns and DNUPs. Third, feature extraction is carried out under the GCLBP framework. Finally, histogram matching is efficiently accomplished by simple nearest neighbor classifier (NNC). The classification accuracy and time-efficiency are verified on a widely recognized texture database (Outex) and a real-world steel surface defect database (NEU). The experimental results promise that the proposed method can be widely applied in online AOI instruments for hot-rolled strip steel.Index Terms-Surface defects, image classification, hot-rolled strips, local binary patterns (LBP), automatic optical inspection (AOI) instrument.

Research Progress of Automated Visual Surface Defect Detection for Industrial Metal Planar Materials

“…It cannot be ignored that the useful description information in the non-uniform patterns in all these LBP variants has been elided. Luo et al creatively used the reverse thinking to explore the non-uniform pattern to supplement the description information hidden in the uniform pattern in [ 7 ] and [ 61 ]. As lightweight feature descriptors, LBP and its variants can be applied to defect detection and classification at the same time, and researchers should follow and develop LBP variants or LBP-like descriptors with better noise robustness and scale invariance, which is also in line with the future development trend of AVI.…”

“…However, the computer-vision-based surface defect detection methods are the most commonly used to find and locate the abnormal areas on the image surface due to their advantages of low cost, easy operation, and superior performance, etc. Nowadays, with the rapid development of hardware facilities and the continuous advance of artificial intelligence technology, automated visual inspection (AVI) equipment has gradually become the standard configuration for industrial manufacturers to improve product quality and production efficiency [ 3 , 4 , 5 , 6 , 7 , 8 ].…”

Automated Visual Defect Detection for Flat Steel Surface: A Survey

“…Nevertheless, descriptive information among non-uniform patterns have been ignored in all these LBP variants. Using reverse thinking, Luo et al [3] proposed a generalized completed local binary patterns (GCLBP) by first exploring the non-uniform patterns to supplement the descriptive information in uniform patterns. Further the work of GCLBP, Luo et al developed a more effective LBP-descriptor (namely SDLBP) in [46], which has remarkable advantages in anti-interference and simplicity of calculation.…”

“…A general AVI instrument provides two main functions of defect detection and classification [1][2][3][4]. The former detection process recognizes defective regions from normal background without identifying what types of defects they are.…”

“…Continuous casting and rolling production lines involve multiple sufferings of high temperature, dense mist, heavy cooling water drops [5], uneven illumination, stochastic noises [1,2], and aperiodic vibration [6]. The undesirable image quality requires preeminent detection algorithms to resist large intra-class variation and minor inter-class distance [1][2][3][4]. 2) Eternally continuous image streams.…”