Ceramic Microstructure Image Segmentation By Mean Shift
Cai Huahui
1, a
, Cheng Yan
2
and Liu BingXiang
1
1
School of Information Engineering, Jingdezhen Ceramic Institute, China
2
School of Art & Design, Jingdezhen Ceramic Institute, China
a
huahuicai@gmail.com
Keywords: Ceramic Microstructure, Median Filter, Image Segmentation, Mean Shift.
Abstract. In order to effectively assist the researchers conduct quantitative analysis of ceramic
microstructures, a segmentation algorithm based on mean shift is used for the ceramic microstructure
image. Since the collection and transfer process of microscopic image will inevitably be subject to
uneven distribution of light, electronic noise and other interference factors which make the image
quality deterioration, it is necessary to reduce noises and enhance edges for ceramic microscopic
image processing at first. Therefore, the median filter is used to remove the noises in the ceramic
microstructure images. Then the component with similar feature is separated and merged by the mean
shift segmentation algorithm. Experiments show the proposed algorithm of using median filter and
mean shift clustering gives satisfactory results.
Introduction
The properties of ceramic are strongly dependent on its microstructure, so the microstructure is the
basic investigation in the science and engineering of ceramic materials [1]. Ceramic microstructure
can be observed through the scanning electron microscope, which contains phases presence and
distribution, grain boundary character distributions, etc [2]. Traditionally, the detection of ceramic
microstructure image mainly rely on human, so it is tedious and time-consuming, more importantly,
the reliability and validity of test results is difficult to guarantee. Since digital image process
technology can effectively help researchers more in-depth structural analysis and research. Therefore,
in order to help researchers for the further quantitative analysis and the measurement, it developed
gradually by the use of scanning electron microscopy and digital image processing technology to
study the characteristic of ceramic particles.
Image Segmentation is one of the most important concerns in digital image processing and a long
standing problem in computer vision [3]. The acquisition and transmission process of the ceramic
microscopic image will inevitably be affected by interferential information, such as electronic noise,
uneven illumination. Thus in many cases the traditional methods of image segmentation and edge
detection were difficult to obtain satisfactory results to the ceramic microscopic image. Reference [4]
first use median filter to remove noise, and then use region growing method for image Segmentation,
Reference [5] first use curvelef transform to remove noise, and then use watershed algorithm for
image Segmentation. However, experiment show that the two segmentation algorithm may be
ineffective although the ceramic microstructure image had been filtered to reduce the noise.
Mean-shift is a non-parametric feature-space analysis technique, application domain include
clustering in computer vision and image processing. It was originally presented in 1975 by Fukunaga
and Hostetler[6], and was improved and given possible applications in 1995 by Yizong Cheng [7],
and more extended to low-level vision problems, including, segmentation [8], tracking[9], etc.
In this paper, it is first discussed median filter and mean shift for image segment. Then, ceramic
microstructure images were segmented by the proposed algorithm of using median filter and mean
shift clustering, the segmentation results for ceramic microstructure images are verified by mean shift
algorithm, and the example shows the usefulness of the algorithm in quantitative analysis.
Applied Mechanics and Materials Vols. 423-426 (2013) pp 2602-2605
Online available since 2013/Sep/27 at www.scientific.net
© (2013) Trans Tech Publications, Switzerland
doi:10.4028/www.scientific.net/AMM.423-426.2602
All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of TTP,
www.ttp.net. (ID: 98.158.109.170-15/02/15,05:51:09)