This paper deals with the contribution of distance maps to edge detection and object rebuilding on grey-level images. Several different distance maps have to be computed first to obtain good results about edge detection, and second to rebuild each particle. Before distance computaion it is necessary to determine the binary images which will be evaluated on. A few classical methods are used to get these images, such as thresholding, gradient computation and so on. For edge detection, two different maps are comsidered. The first step is to compute a good threshold allowing to obtain two classes of points, one for particles and the other for background. As the particle density is very high on studied images (particles can even cover each other), thresholding by itself does not give good results. The second step is to compute the gradien image associated with the initial image. Filtering this gradient image in order to determine the particle boundaries is a difficult problem to solve. Classical methods such as thresholding and elimination of small objects have been used but they do not give results good enough by themselves. Distance maps will be used to improve these results. Then, a combination of two distance maps, one being computed from the thresholded initial image and the other from a particle map, enables this filtering. The particle map is obtained by adding the thresholded image to remaining points by a high-pass filter on gradient values (these points are taken into account as background poins on the particle map). The particle map and the thresholded image are binary images and a distance map can be computed on each of them. Both distance maps are then correlated in order to decide whether a point is noisy or not. This method allows to disconnect most of the particles. This algorithm has been implemented for an application in civil engineering. The aim was to determine the granulometry of a riprap (set of big stones covering an embankment) by image processing. Stone diameters are then computed on the particle map after noise elimination. The second problem is to rebuild particles in order to compute their Féret diameter. One more, two distance ,maps are used. One giving a marker of each stone and second a print of the stone. The whole image is then rebuilt particle by particle and the knowlwdge of particle edges enables the Féret diameter computation.
distance map, gradient filtering, granulometry, particle disconnection, particle rebuilding.
All articles published by Studies in Informatics and Control (SIC) are published in accordance with the terms and conditions of the Creative Commons Attribution License (CC BY-NC-SA 4.0).