ECOCEAN Whale Shark Forum
May 18, 2012, 07:38:33 AM *
Welcome, Guest. Please login or register.

Login with username, password and session length
News:
 
   Home   Help Search Login Register  
Pages: [1]
  Print  
Author Topic: Pre-processing an image for optimal pattern recognition using two algorithms  (Read 1387 times)
Jason Holmberg
Administrator
Newbie
*****
Posts: 113


View Profile WWW
« on: February 17, 2007, 04:33:45 PM »

I'm reproducing here a section of the ECOCEAN Field Station Manual v2 related to pre-processing an image before spot extraction in order to start a discussion on standardized data processing to:

-minimize the likelihood of misidentification by using dual pattern recognition algorithms
-ensure processed images are easy to visually compare after the algorithms have been run by focusing only on the fiducial region and the remapping of matched spots to it
-promote the development of comparable images and digital patterns for those not using an online system, such as the ECOCEAN Library

This process adds a few steps many of you may not currently be incorporating into your own approach. We've found that images generated by this process are easy to visually compare and are compatible with the requirements of the I3S and modified Groth algorithms.

Let's discuss.

---------------------------------------------------
Any encounter that is not assigned to a shark can have new left and/or right side spot data added to it in the ECOCEAN Library if properly oriented photos were submitted for it. Here are the steps required to extract a pattern from a photo for optimal pattern recognition using the I3S and modified Groth algorithms. Let’s use the following photo as a good example.



Figure 1. A new photograph submitted to the ECOCEAN Library.

Pre-processing an image in Paint.NET

If you choose to use the free Paint.NET application to pre-process source images on Windows XP/Vista, use the following instructions. Alternative instructions for using Adobe Fireworks (Windows/Mac) can be found in Appendix D of the ECOCEAN Field Station Manual version 2.0. If you're using a Linux system, the software package GIMP offers similar functionality.

Paint .NET can be freely downloaded from:
http://fileforum.betanews.com/detail/PaintNET/1096481993/1
Paint.NET also requires installation of the Microsoft .NET Framework 2.0:
http://www.microsoft.com/downloads/details.aspx?familyid=0856eacb-4362-4b0d-8edd-aab15c5e04f5&displaylang=en



Follow these steps to pre-process an image:
1. Open Paint.NET on your computer.
2. Open the picture in Paint.NET by selecting Open from the File menu.


Figure 2. A whale shark image loaded in Paint.NET.

3. Use the Rectangle Select tool to select the spot pattern area used for whale sharks.
 

Figure 3.  Selecting the spot pattern area in Paint.NET.

4. From the Image menu, select Crop to Selection to reduce the image to only the needed patterning area.


Figure 4.  A cropped spot pattern area.

5. We also recommend reducing the size of very large images to ensure the spot pattern area fits neatly onto the Paint.NET canvas at 100% magnification. To adjust image size, use the Rectangle Select tool to select the entire image and then select Resize from the Image menu. Adjust image size appropriately.
6. After obtaining a reduced image of only the spot pattern area, create a new layer to hold a horizontal adjustment line using the Add New Layer button of the Layers pallet.


Figure 5.  Adding a new layer to hold a reference line.

7. With the new layer selected in the Layers palette, add a horizontal reference line to the image using the Line\Curve tool . Hold the Shift key when drawing the reference line to ensure it is perfectly horizontal.


Figure 6.  Adding a horizontal reference line.

8. Select the Background layer containing the cropped spot patterning area and then use the Rectangle Select tool to select the entire patterning area image. From the Layers menu, select Rotate/Zoom.


Figure 7.  Adding a horizontal reference line.

9. Use the Rotate and Pan controls of the Rotate/Zoom dialog box to rotate and move the spot patterning area until the vertebral column is flat against the horizontal line. Click OK when you are done.

Note: The section of the vertebral column just above the fifth gill may curve downward slightly and not fit flush to the line. This is acceptable so long as the rest of the vertebral column above the pectoral fin is parallel to the line.


Figure 8.  Obtaining a proper orientation for the spot patterning area.

10. Select the layer with the horizontal red line in the Layers pallet and click the Delete Layer button to remove it.
11. Use the Rectangle Select tool and the Image, Crop to Selection menu command to reduce you image to only the needed spot patterning area.
12. Choose File, Save As to save your completed processed image under a new name.


Figure 9.  An image ready for spot pattern extraction.

You are now ready to extract the spots using I3S or Interconnect. Matched spots can easily be remapped to images such as these, focusing the eye on only the fiducial region and the results of scans for matches.


Figure 10.  A match to H-019 made with the modified Groth algorithm. Potential spot pairs are marked with the same color in both images.
« Last Edit: February 27, 2007, 12:50:55 AM by Jason Holmberg » Logged
Pages: [1]
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.11 | SMF © 2006-2009, Simple Machines LLC Valid XHTML 1.0! Valid CSS!