The proposed method clearly brings solutions to improve the diagnosis on eddy current images. The definition of a validation domain allows to have a threshold solution and an assessment of the solution. From these results, it is easy to define an automatic method to process eddy current images. The images whose results are out of the validation domain could be presented to the expert with a warning comment, others are proposed for validation or even completely automatically thresholded. This thresholding method, through several works on ultrasonic images and through lastest works on eddy current images, has now reached an interesting stage. A clear model of images which can be potentially processed by this method is defined. Automatic analysis of the quality of the solution is possible thanks to the two quality indicators: Difference and Confidence degree. The choice of vector d is a quite easy task for even a new trained operator or it can be automatized by characterization curve. An extension of this work could be a complete integration in an A.I. system dedicated to NDT images analysis. Applications to radiographic or radioscopic images are possible. The research of a more local threshold to process images with a changing average value is also possible but may be blocked by the obligation of a difference in size between defects and image's areas. Other ways to separate the background core in the co-occurrence matrix could improve the cooperation method.
展开▼
