image science : imagic-5 (otolith) 2-d shape analysis

Software for Otolith Shape Analysis

	Contents:
	Basic Concepts
	Imaging and Sample Data Input Form (SDI Form)
	Converting Images to IMAGIC-5 Files
	Obtaining Shape Descriptors
	Reconstruction of Shapes from Shape Descriptors
	Multivariate Statistical Analysis (MSA)
	Hierarchical classification
	IMAGIC-5 Terms

IMAGIC-5 Terms

A fully functional but currently not yet official Windows 95/98/NT4 version is available. The release of the official IMAGIC Windows version will come pretty soon and will feature a full fledged Graphical User Interface (GUI). Please refer to our other web pages.

In connection with the Otolith Shape Analysis, we offer a special IMAGIC-5 package that will cover standard IMAGIC-5 applications as well as MSA, and the specific modules as described in this web page. Furthermore included is the Sample Data Input Form (SDI Form) for Windows 98/NT. Licences have unlimited duration and include software installation on the target computer, if appropriate. Also, on request, a complete set of IMAGIC-5 manuals may be obtained free of charge. Of course you can obtain the software alone or in connection with our prefered 'Complete Solution'.

Otolith starting page

Basic Concepts

Image Science, IMAGIC, and Otolith Shape Analysis

Image Science has a strong background in electron microscopical image processing. Please visit our other web pages for more information on special topics.

Although the vast suite of programs compiled under the name IMAGIC-5 was initially developed for very special purpose, its applicability goes well beyond only electron microscopy. It was therefore possible to extend the software philosophy of IMAGIC-5 to yet another field, namely otolith shape analysis.

The strongholds of IMAGIC-5 are its own multiple-image file format and its multiple image processing capability. These features do ideally harmonise with Multivariate Statistical Analysis on large sets of images.

By integration of Otolith Shape Analysis into the IMAGIC-5 environment, all these features open up quite naturally. In the following, it will be shown in chronological order how the individual steps from otolith imaging to the final image classification of large data-sets can be performed. In the following, specific tools for Fish Otolith Shape Analysis will get the highest attention, whereas standard IMAGIC-5 matters will somehow receive less focus.

It should be noted that Image Science is involved in an international project supported by the European Community: Fish Ageing by Otolith Shape Analysis, FABOSA.

Otolith starting page

Imaging and Sample Data Input Form

Imaging is of course the first chapter of the story and without images there would be no image processing. However, as it is important to obtain proper images, it is as important to document descriptions of the sample. The sample data must therefore accompany each single image.

The Sample Data Input Form (SDI Form) makes entering important information about investigated fish otolith samples and handling of the camera an easy thing. You may take a look at a screen shot of the SDI Form window. As you can see, most input fields are marked with self-explanatory labels. Also, a lot of entry fields have easy-to-use pull-down menus. Not only that the pull-down menu items can be customised by the user, they can be optimised for different tasks and stored in a disk file which is then selected as needed.

Camera initialisation and image acquisition can be arranged from within the SDI Form (buttons: Initialise Camera, Start Imaging, and Stop Camera). The camera software-module itself is by LEICA. It simplifies the use of the camera in a nice way. Have a look at a screenshot of the active camera image acquisition window. After the image is taken and saved, you can minimise the camera window and return back to the SDI Form.

Otolith starting page

Converting Images to IMAGIC-5 Files

The IMAGIC-5 image converter is called EM2EM. It is a special command that allows conversion of image formats used in 3DEM (electron microscopy) all over the world.

We added a special purpose format called 'FABOSA' for input of fish otolith images in TIF format together with any associated sample data text files. The output file is a fully qualified IMAGIC-5 image file that additionally contains the sample data. It should be noted that batch conversion of images can be managed either by explicitly specifying a list of file names on file or by chosing a range of numbers that refer to a numbering portion of the input file names.

Otolith starting page

Obtaining Shape Descriptors

We start out with the grey scale images stored in an IMAGIC-5 file.

Each image is binarised and subjected to sieve filtering yielding a black-and-white image that contains a single object. Edge finding is applied to obtain the contour line which is subsequently spike-filtered. The closed-contour is finally subjected to Discrete Fast Fourier Transform from which the associated Fourier Descriptors are obtained.

An orientation and size normalisation is then performed. An example is shown in the figure. It is worth to note that you can interactively follow and influence the processing of the input image into the contour.

Analysis of a single image takes 1-3 seconds meaning that you can relax for 3-5 minutes while waiting for 100 images.

It should be noted that due to the data processing a striking data reduction from typically 768*582*3 Bytes to only 1300 Bytes is achieved. For larger data-sets it means hundreds of kilobytes only instead of hundreds of megabytes!

There is a diversity of different parameters stored for each original sample. When processing files with IMAGIC-5 you don't have to worry about these parameters, e.g. when appending some entries on one file to another. IMAGIC-5 even maintains a header section for the file history.

Although you don't have to worry about shape and other parameters, there might be a situation where you would like to inspect each single parameter stored. With the command PRINT-NFD-INFO you get a standard text file output which contains all relevant parameter data for each of the images in readable form.

Otolith starting page

Reconstruction of Shapes from Shape Descriptors

GENERATE-IMAGE-FROM-NFD is a command to generate 2D images from Normalised Fourier Descriptors (NFDs). It has the following options:

	Background: black or white
	Draw Style: draw only contour line or filled objects
	Size of Objects: - small reproduction - draw all objects at the same scale - reproduce original size
	Orientation of Objects: - reproduce original orientation - show objects in normalised orientation

One other tool should be mentioned in the context of recalculated shapes. This self-explaining command has the name FLIP-NFD-UPSIDE-DOWN. It should be noted that 'flipping a set of NFDs upside down' is a symmetry relation applied to the Fourier coefficients rather than the usual rotation applied to the pixels of an image. After the symmetry operation, the recalculated objects appear to be rotated by 180 degrees.

Otolith starting page

Multivariate Statistical Analysis

A number of commands for Multivariate Statistical Analysis (MSA) are available in the IMAGIC-5 package. Here, only some connections to fish shape analysis are indicated.

	small reproductions of the normalised objects are generated as input for MSA
	eigenimages are determined
	factorial co-ordinates in the eigenimage system are calculated
	classifications based on factorial co-ordinates are performed
	more ...

As described earlier, the Normalised Fourier Descriptors are parameters for shape classifications. The above procedure is therefore an alternative approach. In most applications factorial co-ordinates and Fourier Descriptors are both equally well suited for classification purposes.

Otolith starting page

Hierarchical classification

There are two more tools that should be mentioned in connection with classifications. These tools concern the extraction of the members of a class into one separate file and calculating the class average. In fact, IMAGIC-5 offers two strategies to perform these tasks. One strategy uses reproduced shapes, the other strategy uses NFDs:

MSA-EXTRACT-SORTED-CLASS
MSA-SUM-CLASSES
MSA-SELECT-NFD
MSA-AVERAGE-NFD

The IMAGIC-5 command MSA-CLASSIFY denotes a sophisticated instrument for hierarchical classification (HAC). The original HAC is refined by the Moving-Elements method, which generally puts the original HAC close to the theoretical optimum. It is worth to note that these tools were especially devised to cope with large data-sets of several ten-thousands of objects.

When dealing with medium size data-sets, a graphical representation of the hierarchical classification tree is very helpful for interpretation of the inner data structure. A plot of the HAC tree can guide you, e.g. to find the number of classes into which the data-set should be subdivided.

Otolith starting page

Background
Hardware
Technical Support