Skeletonization is a method for shape recognition and characterization that uses discrete geometry notions. Given a binary image, this method allows the extraction of a skeleton.This skeleton is computed thanks to several scans of the image. It is a centered subset of the shape, thin, topologically equivalent to the input shape and it can be reconstructed. In this thesis we aimed at computing the skeleton with only one scan (streaming mode). We thus proposed a new algorithm that re-organizes the computation on each pixel according to their occurrence order within the image in order to compute an asymmetric distance map and to find the medial axis points. The method does not change the direction of the image information propagation. According to a particular order, which guarantees the same results than those of the reference algorithm results, we can delete the points from the asymmetric distance map while conserving the skeletal points and the medial axis points. Besides, the fact that our skeletons are identical to the pixel to those obtained by the reference method, our algorithm is the fastest. Although our approach is generic, we used this algorithm to develop a software tool for medical imaging applications. Trabecular bones characteristics extraction from 2D binarized images enables bones characterization that matches the observations yielded from other software
