文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > 计算机软件及应用 > An Efficient Method for Color Image Segmentation
侵权投诉

An Efficient Method for Color Image Segmentation

图像分割、彩色图像、meanshift

Image segmentation based on neural networks is a promising research area. Neural Network segmentation rely on processing small areas of an image using an artificial neural network [9] or a set of neural networks. After such processing, the decision-making mechanism marks the areas of an image according to the category recognized by the neural network. A type of network designed especially for this purpose is the Kohonen map. Pulse-Coupled Neural Networks (PCNNs) are neural models proposed by modeling a cat’s II.ADAPTIVEMEANSHIFTALGORITHM

It is an unsupervised segmentation algorithm for color images based on Gaussian Mixture models [22]. The number of mixture components is determined automatically by adaptive mean shift, in which local clusters are estimated by repeatedly

An Efficient Method for Color Image Segmentation

visual cortex and developed for high-performance biometric image processing. Eckhorn [28] proposed a neural model to emulate the mechanism of cat’s visual cortex. The Eckhorn model provided a simple and effective tool for studying small mammal’s visual cortex, and was soon recognized as having significant application potential in image processing. The Eckhorn model was adapted to be an image processing algorithm by Johnson who termed this algorithm as Pulse Coupled Neural Network.

Wenbing Tao, Hai Jin [13] proposed two methods to segment the image. These approaches are actually used for the real time applications. In this method the image is first segmented using mean shift algorithm [19][23]. Segments generated after this algorithm are represented as vertex in the graph. Then find the dissimilarity matrix between the nodes. The dissimilarity matrix is determined using the color differences in LUV color space. Key characteristics of the mean-shift algorithm include the fact that no initial cluster positions are required and the final number of extracted clusters is a result of the algorithm. Mean-shift has been widely used for the application areas like image segmentation, object tracking[25] and medical image analysis application [20] [21] [22][26].

Adaptive gradient ascent is used to detect local maxima of data density in feature space. Data points are associated with local maxima[27] or modes, there by defining the clusters. Clustering with mean shift is obtained by initiating the iterative procedure from each feature vector of the given dataset in a d-dimensional space and associating it with the corresponding convergence mode. Usually the number of convergence modes is much smaller than the dataset, with many feature vectors converging to the same mode, thus defining a cluster. In order to define mean-shift clustering it is necessary to choose a kernel and assign the values of the window size hi. .D. Comaniciu et.al [24] proposed the properties of different kernels, including convergence conditions, smoothness of convergence trajectory and other properties.

Section II describes the Adaptive Mean Shift algorithm and the Graph Partitioning method. The proposed method, combined approach of Adaptive Mean Shift algorithm with Normalized Cuts is discussed in Section III. Result analysis and the performance comparison of the proposed method is explained in Section IV. Section V focuses on Conclusion.

978-1-4673-0671-3/11/$26.00©2011 IEEE kernel, and the normalization constant k,d

Assumes that K(x) integrates to one. The function g(x)=-k|(x) can always be defined when the derivative of the kernel profile k(x) exists. Using g(x) as the profile, the kernel G(x) is defined as G(x) =C2

g,dgx

.

By taking the Gradient of (1), the following property can be proven.

mG(x) = C

fk(x)

f

G(x)

(3)

where C is a positive constant and

n

1

§x x2¦dxg¨i·

¸im 1h 2ii

¨hi¸G(x) =

¦n x (4) 1§x x2

·hd2g¨i¸i 1i

¨©hi¸¹

is called the mean shift vector. The expression (3)

shows that at location x the weighted mean of the data points selected with kernel G is proportional to the normalized density gradient estimate obtained with kernel k. the mean shift vector thus points toward the direction of maximum increase in the density. The implication of the mean shift property is that the iterative procedure.

¦n

x§y2

i

j xi

·dg¨¸i 1h2¨h¸

yi

i

j 1=

2¦n

1§¨yj xi

· j=1, 2..

(5)

hd 2g¸i 1i

¨©

hi

¸¹

310

IEEE-ICoAC 2011

上一页第2页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email