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

An Efficient Method for Color Image Segmentation using Adaptive Mean Shift and Normalized Cuts

Shibu V S Member IEEE

Department of Computer Science University of Kerala, Kariavattom, Thiruvananthapuram, Kerala, India

shibuviswam@gmail.com

Philomina Simon Member IEEE

Department of Computer Science University of Kerala, Kariavattom, Thiruvananthapuram, Kerala, India

philomina.simon@gmail.com

Abstract—In the proposed method, a combined approach of Adaptive Mean Shift and Normalized Cuts is used for clustering the images. In this method, both color and gray scale images can be segmented effectively and it requires less computational complexity. In the first stage, the image is divided into different segments using Adaptive Mean Shift algorithm and the segments generated are labeled and the labeled segments are represented as nodes in a graph. The result obtained by applying the Adaptive Mean Shift algorithm is given to the normalized cut method for grouping the clustered segments. Experimental result shows that the proposed method gives better performance in terms of segments than other methods when tested with color and gray scale natural images.

Keywords-Adaptive Mean shift, Normalized cut, Image Segmentation , Clustering

I.INTRODUCTION

Image segmentation divides the digital image into multiple regions or groups. Each group is distinguished by other groups based on their intensity levels, color, texture or other properties. Color image segmentation is a bit more complex compared to gray level images. Color images contain multiple layers. Thus image partitioning becomes harder. Image segmentation separates the background from the objects in the image. Adjacent regions can be separated based on the different features extracted such as color, texture, region and edges. Color image segmentation is used for the research areas like computer vision, medical image processing, robotics, Object tracking in satellite images, Face recognition, and Traffic control systems. Image segmentation techniques are classified into Thresholding methods [10][11][12], Region growing method [1][3][15][16][17], Graph partitioning method[4] [5] [6] [7] [8][14], Histogram based method[1] [2], Neural network based methods[9].

Thresholding methods are the simplest method among all the segmentation techniques. In this method the given image is segmented into different regions using a threshold value. Thresholding methods look for the boundary discontinuities. The key point in this method is the selection of threshold value. Otsu’s method is one of the approaches to find out the

threshold value. The region growing methods are based on the intensity values. The first step in region growing is to select a set of seed points. Seed point selection is based on some user criterion such as pixels in a certain gray level range, pixels evenly spaced on a grid. The initial region begins as the exact location of these seeds. The regions are then grown from these seed points to adjacent points depending on a region membership criterion. The criterion could be pixel intensity, gray level texture or color. One of the challenging task in region growing is selection of seed points. Limitations of region growing include time consumption; noise or variation of intensity may result in holes or over segmentation and this method does not differentiate the shade part of the real images. A classical technique for color image segmentation is the K-means (or C-means) algorithm [29], widely adopted also for vector quantization and data compression. An original technique, proposed by Yung and Lai[30], adopts the constrained gravitational clustering: two points xi and xj within the RGB pattern space are modeled as two particles Pi and Pjrespectively having masses mi and mj, and interacting

according to the gravitational law

F Gmimj(xi xj)/xi xj,where G is the

gravitational constant.

In Graph partitioning methods [4] [5] [6] [7] [8][14], the image is modeled as a weighted, undirected graph. Usually a pixel or a group of pixels are associated with nodes and edge weights define the dissimilarity or similarity between the neighborhood pixels. The graph is then partitioned according to a criterion designed to model good clusters. Each partition of the nodes (pixels) output from these algorithms are considered an object segment in the image. Commonly used algorithms are minimum cut [5], normalized cut [4], spanning tree-based segmentation [6], isoperimetric partitioning [7]. Limitation of the method includes huge computational complexity. In histogram based approach, histogram is generated from all the pixels in the image. The regions can be distinguished using the peaks and valleys in the histogram. But it is very difficult to find the significant valleys and peaks in the histogram.

3

978-1-4673-0671-3/11/$26.00©2011 IEEE 309

IEEE-ICoAC 2011

第1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