2 - Matrice de co-occurrence optimale pour la segmentation automatique d'images ultrasonores

This paper introduces a new method of segmentation using automatic thresholding adapted to the NDT ultrasonic images . This study is based on image analysis through co-occurrence matrixes . It shows an optimization of the r and 0 parameters of the co-occurrence matrix enabling to define more acurately the border between noise and defect echoes . The segmentation is obtained by automatically taking into account a threshold derived from a determination curve calculated front the co-occurrence matrix . This curve, called Average Product of Variances Measure, is an analysis of the distribution of the matrix coefficients . The results show behaviors of the co-occurrence matrixes and of the threshold selection curves that justify perfectly the analysis performed on the characteristics of the image .