Bscan image segmentation by thresholding using cooccurrence matrix analysis

摘要

In this paper we present innovative research for the processing of an uktrasonic image obtained in non-destructive testing (NDT). We also present an algorithm for automatic thresholding segmentation. The image analysis is based on the calculation of cooccurrence matrices. The method we used is the Spatial Grey-Level Dependence Method (SGLDM); it shows the distribution of the intensities of pairs of pixels separated by a vector d. We show that it is necessary to build optimized matrices in order to improve the existing techniques while adapting to the specifics of the ultrasonic image obtained in NDT. We solve for the vector d through an automatic analysis of the matrix behaviour for several vectors based on the characterization curves. The developed method allows us to present an efficient thresholding algorithm that is tested on real images.