Defect Inspection Based on Segmentation and Defective Tracking in Welding Images with Orthogonal Polynomials

Abstract

Welding image defect tracking is the process of compensating missing details in newlinewelding image regions or repairing damaged portions in a digital welding image. It aims to newlinethe damaged or missing part of the welding image by collecting low level information from newlineundamaged part, in such a way that the defect part is natural and complete. In recent years, newlinedefect tracking in digital welding images has gained significant attraction, due to its need in newlinevariety of applications.. There are two generic approaches to extract the low-level primitives newlinein undamaged portions: Partial Differential Equation (PDE) based anddefect texture synthesis newlinebased. The partial differential equation based schemes result in blurring artifacts, when the newlinedamaged or missing area is larger in size. The defect texture synthesis based defect tracking newlineschemes can overcome this difficulty and work well for large target region, simultaneously newlinepresenting linear structure information. But it cannot handle effectively the curved structures. newlineHence in this research work, new welding image defect tracking techniques that can identify newlinethe defect area containing both structure and defect texture information with orthogonal newlinepolynomials model is proposed. newlineThis research work, initially proposes a generic defect detection system with newlineorthogonal polynomials transcoded coefficients. In this direction, the proposed defect newlinedetection system partitions the welding image into blocks and each block is applied with newlineorthogonal polynomials transform, and subjected to a modified lifting scheme. The resulting newlinetranscoded coefficients are modeled as a probability density function and propose a new newlineblock classification scheme, with less computational complexity. Defects in edges blocks are newlinethen identified with a group distribution model. While the proposed defect detection system newlineidentifies the defects in smooth area with simple comparison of magnitude of transcoded newlinecoefficients, texture defects are identified by employing homogeneity among variances of newlinetranscoded coefficients.