sem analysis

9
Software for the Image Analysis of Cheese Microstructure from SEM Imagery Gaetano Impoco impoco @ dmi.unict.it September 12, 2007 1 Contents This software is intended for the analysis of SEM 1 imagery of cheese mi- crostructure. It might be also used in connection with different acquisition technologies or for different applications where porous materials are involved. However, being tailored to a specific application, you cannot be sure it will be of any use in your application. Applicability of this software should be evaluated for every specific application. This software is in the form of a plug-in for ImageJ [3]. ImageJ is a widely used Open Source software in scientific communities employing Image Analysis and is released under the GPL license [1]. Figure 1 shows a snapshot of ImageJ and the plug-in for the analysis of cheese SEM imagery. The plug- in encompasses two commands: BinariseSEM ComputeStats. BinariseSEM segments 2 the input image into “holes” and “structure” i.e., in areas that are interpreted as pores or as protein matrix. It is useful to remark that: 1 Scanning Electron Microscope. 2 To partition an image into a number of classes, using a coherence criterion within each class. 1

Upload: jing6sha

Post on 06-Feb-2016

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Sem Analysis

Software for the Image Analysis of CheeseMicrostructure from SEM Imagery

Gaetano Impocoimpoco @ dmi.unict.it

September 12, 2007

1 Contents

This software is intended for the analysis of SEM 1 imagery of cheese mi-crostructure. It might be also used in connection with different acquisitiontechnologies or for different applications where porous materials are involved.However, being tailored to a specific application, you cannot be sure it willbe of any use in your application. Applicability of this software should beevaluated for every specific application.

This software is in the form of a plug-in for ImageJ [3]. ImageJ is awidely used Open Source software in scientific communities employing ImageAnalysis and is released under the GPL license [1]. Figure 1 shows a snapshotof ImageJ and the plug-in for the analysis of cheese SEM imagery. The plug-in encompasses two commands:

• BinariseSEM

• ComputeStats.

BinariseSEM segments 2 the input image into “holes” and “structure” i.e.,in areas that are interpreted as pores or as protein matrix. It is useful toremark that:

1Scanning Electron Microscope.2To partition an image into a number of classes, using a coherence criterion within each

class.

1

Page 2: Sem Analysis

Figure 1: A snapshot of the ImageJ software and the plug-in for ImageAnalysis.

2

Page 3: Sem Analysis

• Holes and structure that are not imaged (i.e., acquired by the micro-scope) cannot be processed by no Image Processing software, simplybecause they are not visible.

• Pores with diameter of one pixel o less are discarded since they aretreated as acquisition noise.

• The definition of pore used here does not necessarily coincide with theinformal definition nor with the expectations of the user. In particular,here a pore is defined as a set of contiguous pixels whose intensityvalues are coherent. The coherency function derives from the sequenceof operations used to enhance the input image before thresholding 3)

The command ComputeStats employs the output of the previous appli-cation BinariseSEM to collect image statistics about the distribution of pores,such as: number of pores, perimeter and area, shape descriptors, orientation,and so on.

It is assumed that input images are achromatic (i.e., greylevel) and witha bitdepth of 8-16 bit/pixel.

2 Installation

The installation procedure of ImageJ plug-ins is quite simple. Just extractthe content of the downloaded compressed file into the directory [Image-JbaseDir]/ plugins/ where [ImageJbaseDir] is the ImageJ installation direc-tory, and run ImageJ.

3 The Plug-in

The SEM analysis plug-in has two commands that must be used in the cor-rect sequence. ComputeStats takes a binary image as input, where it isassumed that white pixels belong to holes and black pixels represent struc-ture. BinariseSEM outputs this kind of images, given a greylevel SEMimage.

3Image Processing operation used to segment the image using a threshold value. Thisterm is mainly used for binary segmentation (i.e., with two classes only).

3

Page 4: Sem Analysis

We chose to break the plug-in into two commands in order to allow easysubstitution of BinariseSEM with another binarisation software giving thesame output (for example, a simple global thresholding). A more detailedpresentation of the two commands follows. For a description of the algorithmsused the reader is referred to [2].

3.1 BinariseSEM

This command segments the input image into pores and structure, where apore is defined as a patch of contiguous pixels with similar intensity values.A discontinuity in the pixel intensity is regarded as a discontinuity in thestructure of the material. This might be not always true, due to noise andreflection effects caused by the acquisition device.

As shown in Figure 3, after opening an image e selecting the Binaris-eSEM command in the menu, a dialog window is shown. Two options aregiven: Bandpass Filter e Fill Inner Holes. It is asked to the user to guidethe segmentation mechanism since these two parameters are strongly depen-dent on the application and on the domain knowledge of the user, and theirvalues cannot be automatically guessed for every application. In particular,the second option lets the user choose wether the holes must be filled or notafter thresholding. Hence, it should be turned on only if the user knows thatthere cannot be any structure inside a hole.

The first option is used to correct possible illumination gradient effectsdue to the acquisition system. When activated, a filter is used to correct forintensity gradients but, at the same time, the image quality is reduced sothat the quality of the output could reduced as well. Hence, care should betaken when using this option. Before using the software for massive analysis,we recommend to run this command twice for each image using this twooptions alternately in order to get the feeling of the result.

After selecting the desired options the command is run. A sequence ofprocessing operations are executed to enhance the quality of the input imageand to make it more amenable for thresholding [2]. When the processing endsa window is shown to help the user choosing the best threshold value for theapplication (Figure 3(a)). The plug-in automatically computes a thresholdvalue that is suggested to the user. Ino most cases, this value gives a goodthresholding. However, an automatic procedure cannot give optimal resultsfor all applications and has no knowledge about the domain of application.Hence, the thresholding procedure must be assisted by the user. When the

4

Page 5: Sem Analysis

(a)

(b)

Figure 2: Options of the BinariseSEM command.

5

Page 6: Sem Analysis

(a) Thresholding of the processed image.

(b) Binary output image.

Figure 3: Thresholding.

6

Page 7: Sem Analysis

threshold value has been chosen the output will appear as in Figure 3(b).

3.2 ComputeStats

This command collects various statistics about the holes extracted using theBinariseSEM and shows them using histograms and rose plots. Figure 4shows the options of this command. The most important is the magnificationfactor of the microscope, expressed in microns. The other options refer tothe statistics to evaluate and show. Figure 5 reports an example of theoutput of this command. Histograms are shown reporting the distribution ofseveral measures and of the value of some shape descriptors (see [2] for a moredetailed description). All the distributions are summarised by histograms.Directionality can be also shown using an angle diagram (rose plot). Onesuch diagram is shown on the bottom right of Figure 5. Notice that thestatistics about directionality report a predominant orientation of the poresapproximately around 45 with respect to the x-axis (see the rose plot andthe histogram on its left).

References

[1] Free Software Foundation. GNU General Puplic License. Web site: http://www.gnu.org/licenses/licenses.html#GPL.

[2] G. Impoco, S. Carrato, M. Caccamo, L. Tuminello, and G. Licitra.Quantitative analysis of cheese microstructure using sem imagery. InSIMAI 2006. Minisymposium on Image Analysis Methods for IndustrialApplication, 2006.

[3] Research Services Branch NIMH & NINDS. ImageJ - Image processingand analysis in Java. Web site: http://rsb.info.nih.gov/ij/.

7

Page 8: Sem Analysis

(a)

(b)

Figure 4: Options of the command ComputeStats.

8

Page 9: Sem Analysis

(a)

(b)

Figure 5: Statistics computed.

9