In order to improve the effectiveness of Snake active model

the deformable template approach of contour-tracking algorithm

in color images

this paper mainly focuses on resolving two problems. First

to reduce the difficulty of handcrafting initial contour for Snake

the authors introduce the pretreatment procedure. They first use color-based clustering algorithm to divide an image into several regions

then use improved graph-searching algorithm to get the contour of independent regions which may be important for further recognition

and take this result as the initial contour. Second

to avoid the distortion contour produced by Snake model when applying it to tracking the contour in some color images

they redesign the external energy function of Snake. They use neighborhood pixels' Euclidean distance of the colors in weighted HSI color space

instead of grey level difference in traditional algorithms

to approximate the grads of image

and take the square of the distance as external energy function to keep the active contour staying at the place where the real contour locates. At last

comparing prototype of Snake model with the new method

authors illustrate that the improvement is considerable

especially when color images are involved.