||In computer vision and pattern recognition, the concept of multi-scale means that data is described or organized considering at least one varying parameter in the spatial or temporal domain. In this talk, I will first focus on the Irregular Isothetic grid model (II-grid model), which permits to generalize many kinds of spatial structurations (quadtrees, MPEG, RLE, etc.), and is a special multi-scale representation. Thanks to this II-grid model, I have also addressed several applications like licence plate recognition or implicit curve approximation. In particular, I will present a recent use of the II-grid model and its related concepts for the vectorization of digital noisy contours. I will finally address the evaluation of segmentation and background modeling algorithms, and raise the problem of robustness in pattern recognition and computer vision.