Statistical detection of defects in radiographic images using an adaptive parametric model

Fully automatic detection of anomalies on wheels surface using an adaptive accurate model and hypothesis testing theory

“…The same applies for the (discrete) coordinates, denoted as x and y, and for the coefficients c i,j of the polynomial (2), denoted as vector c k of (d x +1)×(d y + 1) dimension. Denoting as matrix F the polynomial model, which is detailed in [3,7,8] but not in this paper due to space limitation, the model of the background (2) can be written as:…”

“…The model (1) -(3) is simple and efficient enough for several applications [3,7,8]. However, the non-anomalous background of a wheel is too much complex, with multiple light reflection and artifacts, to be represented with a simple polynomial model.…”

“…It is then easy to establish [12,3] that the norm of the normalized "residuals" r k follows the distribution…”

“…During the past decades computer-vision based non-destructive testing have been broadly used for inspection of wide range of "objects" such as fabrics [1,2], nuclear fuel rods [3] or even food to cite few. However, such systems are usually designed to help and ease an operator who has to do the inspection "manually" which might be subjective and labor intensive.…”

“…. 3) Methods based on computer-vision and image processing [3,4], that usually require prior information on the non-anomalous object. Two main approaches have been proposed to introduce statistical prior knowledge: Bayesian and non-Bayesian approaches.…”