Abstract : Feature extraction is a crucial step in all visual systems. Since a few years, on aims to look for features like interest points, key points or salient points. Local feature point based methods have shown to be very efficient for object recognition. However, points in a high-dimensional feature space do not allow abstract representation of object shape, and, even if they are invariant to imaging conditions, they are not good at generalization. The work represented in this dissertation concerns a very precise question : Which roles can a ridge play for object representation ? We study ridges on surfaces associated to smoothed images in scale-space. Ridge points are directional local extrema, detected by using Laplacian operator. These points are labelled to build ridges lines, useful for object representation. Ridges, by theirs “line” nature, are very useful to represent structural object line human silhouette or
text line.We proposed to modelize human and text using some significant ridges. A human is represented by ridges corresponding to torso and legs of the human. This representation permits to analyze human
movement via his configuration. A text line is modelized by a long ridge at coarse scale corresponding to the center line of the text line and several smaller ridges at finer scale corresponding to character skeletons. Ridges at small scale must not be parallel to the main ridge. This text model is generic for all types of text and independent with text orientation.