snakes 计算机视觉 算法.pdf

Deformable contour models: snakes Slide sources: original from Kristen Grauman; updated by Vittorio Ferrari The grouping problem Goal: move from array of pixel values to a collection of regions, objects, and shapes. Pixels vs. regions By grouping pixels based on Gestalt- inspired attributes, we can map the pixels into a set of image clusters on intensity regions. Each region is consistent according to the features and similarity metric we used to do the clustering. image clusters on color Edges vs. boundaries Edges useful signal to indicate occluding boundaries, shape. Here the raw edge …but quite often boundaries of interest output is not so bad… are fragmented, and we have extra Images
from
D.
Jacobs
 “clutter” edge points. Edges vs. boundaries Given a model of interest, we can overcome some of the missing and noisy edges using fitting techniques.