Abstract III
Abstract
Image segmentation is the process of extracting the interested target from
background, it is a key step from image processing to image analysis and a very
important position in image engineering. Image segmentation has been widely used in
computer vision, pattern recognition and many other areas. The research of image
segmentation has a very important significance.
Due to the image exists uncertainty and complexity, fuzzy clustering analysis can
well describe this problem, so the fuzzy cluster analysis used in image segmentation
has the very good effect. In many of the fuzzy clustering analysis method, the Fuzzy
C-Means (FCM) clustering is the most classic algorithm and has got extensive and
in-depth research. FCM not only avoid the threshold setting, but also very suitable for
processing fuzzy and uncertain problems. And it is a kind of unsupervised clustering
algorithm, do not need to artificial intervention, suitable for automatic segmentation.
Therefore, using FCM clustering algorithm for image segmentation has become one of
the main direction research. In this paper, aim at fuzzy c-means clustering algorithm
exist some problems in image segmentation to improve:
First of all, because the traditional fuzzy c-means clustering algorithm is not good
use of its own image spatial information, leading to it is very sensitive to noise.
According to this problem, we combined with non-local means, using image structure
space information, proposed a new method based on non-local fuzzy c-means image
segmentation algorithm. This algorithm makes full use of the image space information,
effective reducing the influence of noise to the segmentation results. The results of this
method for different types noise under different image show that it can obtain a better
segmentation results.
Secondly, the traditional fuzzy c-means image segmentation algorithm in decided
to each pixel class, only use the difference of each pixel and clustering center, it is
not suit to analysis the neighborhood pixels class mark. Update the clustering center
using all pixels in an image, this will lead to the clustering center of migration. Based
on the above problem, we use the thought of principal component analysis and using
neighborhood information reasonably, to determine each pixel class mark; Then using
the idea of rough set, in every clustering center update, use clear belong to the center
and in the center of edge pixels, and then proposed based on local PCA and rough set
