Image segmentation refers to the process of partitioning a digital image into multiple segments. The goal of segmentation
is to simplify and/or change the representation of an image into something that is more meaningful and easier to analysis
[1]. These advantages of this process make it be widely used to find out the interesting objects in high resolution remote
sensing images. Watershed algorithm is based on the topology of the image, although it can easily split images into
homogenous partitions, it also leads to over-segment in practical implementation. Some solutions were proposed to solve
this problem in the past few years [2][3]. Using pre-defined seeds and extracting the pixels clusters which are grown from
these seeds is a reasonable method to overwhelm the obscure of over-segmentation.
In this paper; we present a novel framework to improve the results of watersheds segmentation by using edge detection
to find out the position of seed points. Then the immersion simulations suggested by Vicent has been used to segment the
image. The algorithm consists of four steps: a) Edge detection with embedded confidence, b) Thinning processing on the
previous results, c) Label the seed points on each side of the thinned edges, d) Detection of watersheds on gradient
magnitude image using immersion simulations.
At last, we use high resolution remote sensing image to qualify the framework and the experimental results are presented.
It shows that reasonable results which preserve the edge could be gotten by applying this framework.
|