Objects can be controlled by different methods, such as radiographic records for example. In this case, the image processing techniques allow to really measure the defects, and not only estimate them. This increasing precision of the results has been obtained by defining a set of correcting parameters in order to take into account experimental constraints, characteristic of the X-Ray generator and linked reception system, and of the use of a self-calibrating operating process. The measurement of the defect(s) can be seen as acted in three parts : a) Preliminary measurements so as to initiate the correcting parameters, activation of the corrections if necessary and if possible, measurements of the calibrating areas. b) Extracting the defective zone from the total image. The classical optimal separator based on bimodal gaussian distribution does not work well, and leads to important errors : the implicit hypothesis of stationarity cannot be held for the defective zone, but the background, or the whole zone, can be regarded as stationary, or varying so slowly to allow correcting it in order to obtain stationarity without modifying the boundary of the defective zone. Our aim is to find out the only pixels belonging to the background, for they constitute a set S, the stati,,tical parameters of which can be known. Pixels, the values of which are lower than a tnreshold Vo, belong to S. Pixels, the values of which are greater than Vo can belong to S, on the condition that a d-distance is lower than a value Vk determined by the number k of pixels already detected as belonging to S in the neighbourhood of the current pixel. The d-distance is based on the first and second statical moments of the S-population. The computation can be iterated and convergence is rapidly obtained. The defective pixels are immediatly obtained as the complementary set S. c) This set S leads to a binary space acting as a mask to be applied on the original image and identifying the pixels belonging to the defective domain. We can, now, measure the characteristics of the defects : maximum height, width, thickness at every location, volume, mass (or lack of mass) ... As an illustration to our algorithm, we give examples obtained from test-samples defined so as to measure the precision one can obtain, and verify the convergence, and results obtained from real defects.