June 3, 2004

Contents

Measurement of Percent Area

This exercise selects features of an image by thresholding, and measures the number of 'red' pixels with the Analyze -> Measure menu.

• Open the Sample Images / MAS Journal Article / TEM Filter Sample.tiff image with the File -> Open menu
  • 
  • (part of image shown).
• Examine the histogram
  • NIH Image only
    • Make sure the Invert Pixel Values box is checked (Options -> Preferences menu), so that lighter image gray levels have higher pixel values.
    • Analyze -> Show Histogram menu.
    • 
    • There are no pixels with low values - corresponding to the absence of darker shades in the image.
  • ImageJ
    • Analyze / Histogram shows this:
    • 
    • There are very few pixels with high values, corresponding to darker shades in the image.

The image appears washed out. It can be enhanced to make the holes in the filter easier to see.

• ImageJ:  Be sure to click on the image to make it active.  Otherwise, the histogram, which is also an image, will be active, and will be "enhanced".  The histogram is primarily a black and white image, and will show no change on "enhancement".
• Process -> Enhance Contrast
  • Use the default settings in the dialog.

The features in the image are easier to see. The gray levels have not been changed, as reapplying the Analyze -> Show Histogram  Analyze -> Histogram menu will show.

• To actually change the pixel values:
  • NIH Image:  Process -> Apply LUT.
  • ImageJ:  Repeat Process / Enhance Contrast, but this time check the Normalize box.

• Analyze -> Show Histogram  Analyze / Histogram to see the modified histogram

  • NIH Image

    ImageJ

  • Note that there are not more gray levels in this enhanced image - the gray levels are just spread out. Adding new gray levels between the old ones might be done with a smoothing operation, but not using the LUT.

Either the enhanced image or the original can be thresholded with the same effect, since the number of gray levels is the same. The enhanced image may be easier to use, since the details are more easily seen, and using the thresholding tool is less tedious.

Measure the percent area of the pores in the filter.

• First, the pores must be selected using the thresholding tool (colored red).
  • NIH Image:
    • Double-click on the thresholding tool , and move the top and bottom of the red band in the LUT window to select the pores as well as possible.
    • 
  • ImageJ
    • Image / Adjust / Threshold Move the sliders to select the pores (turn them red):
    • 

Two problems immediately appear. Red pixels are missing inside the pores where they should be, and there are red pixels outside the pores where they should not be.

• Both problems are almost completely eliminated by one application of the Process -> Smooth menu.

We will ignore the remaining red dots that lie outside the pores. We will also not check for any slight alteration of pore area that the smoothing might have caused.

• Alternative ways to clear up the problems:
  • (Does not apply to ImageJ). The holes inside the pores are due to those pixels having value 255, which cannot be thresholded. These pixels can be selected after adding and subtracting 1 to the image, because of the scaling rules Image uses (see manual).
    • Process -> Arithmetic -> Add, select 1.
    • Process -> Arithmetic -> Subtract, select 1.
  • The extra pixels outside the pores can be taken care of by making a binary image from the thresholded image with the Process -> Binary -> Make Binary Process / Binary / Threshold menu, and then removing the isolated single pixels in the binary image with the Process -> Binary -> Open menu, and analysing the result. The open menu erodes a one pixel layer from every object (black) and then dilates a layer back on (these are available in the Binary menu). Since the erosion eliminates objects that are one pixel thick, there are no pixels left to dilate, and these objects are eliminated, while the others are (almost) their original size. Process / Binary / Close will fill small holes.  Your result might look something like this:
  • 

Counting and sizing pores with NIH Image

• With the thresholded image (Upper and Lower are 227 - 254 in the Info menu, when the mouse button is held down and the threshold tool is in the LUT window)

• Analyze -> Measure menu, see 14127 square pixels in the Info menu.
• Select the entire image with the threshold tool, and use the Analyze -> Measure menu again. See 207944 in the info menu. (Otherwise, determine the area of the image by multiplying the height x width. The only way I know to find that using Image is to carefully move the cursor to the upper right, seeing 499 and 416 for X and Y. The image is 500 x417 - the lower left pixel is (0,0) in Image.)
• The % area (approximately) of the pores is 100 x (14127 / 207944).
• 
• Note that without adding and subtracting 1 from the image (in the bullet above this one), that a small percentage of the area of the particles that has a pixel value of 0 (255), and cannot be selected. We are ignoring this error.
• Measure the percent area of the particles in the image by selecting them. Note that what constitutes 'particle' is your choice - there are faint gray splotches that could be considered particle or background.
• List the pores and their areas and diameters:
  • Select the pores by thresholding as above.
  • Analyze -> Options menu, check Area, Ellipse Major Axis and Ellipse Minor Axis.

• Invoke the Analyze -> Analyze Particles menu, checking all of the boxes, and selecting a reasonable minimum particle size so as not to count any remaining single or double red pixels.
• NOTE: Lables and outlines are drawn right on the image, making it useless for a second analysis. Make a copy of the image, or save it to a file before analyzing.

• An image with the particles labeled and outlined is the result:

• and the properties are listed in a table in a text window using the Analyze -> Show Results menu. This table can be saved as a text file (File -> Save As menu, click the Measurements button) and imported into statistical or plotting packages.
• NOTE: If the Label Particles box is checked, but the particles are not labeled, there are probably more particles in the image than the Max Measurements parameter allows. Increase Max Measurements in the Analyze -> Options dialog and restart Image.

(part of table shown)

Counting and Sizing Pores with ImageJ

This proceedure starts fresh - you might already have the image open from above steps.

• File / Open and select Sample Images / MAS Journal Article / TEM filter sample.tiff.
• Smooth with Process / Filters / Median / radius of 2 pixels.
• Image / Adjust / Threshold, thresholds 249 - 242.
• 
• 
• Analyze / Measure Particles, set the options like this:
• 
• The summary window shows that the percent area of the pores is 6.5%:
• 
• The drawing shows each pore outline and label:
• 
• And the results window lists the area for each pore.  You can cut and paste these results to Excel, or save them to a file.  (Only the first seven pores in the list are shown.)
• 

• Other sample images
  • Sample Images / Proc samples / Apt#3
    • All of the particles can be selected by thresholding. When the darker particles at the bottom of the image are selected, the brighter particles at the top tend to be joined together -- these particles will add to the total a little more than their actual area.
    • To measure the area of the background, which as loaded has a level of 255 (not reachable with the LUT tool)
      • Process->Arithmetic->Add... menu. Select 2 (the default is 25).
      • Process->Arithmetic->Subtract... menu. Select 1.
  • Sample Images / Proc samples / Rainey Nickel.tiff
    • This is more of a challenge because the two lighter phases cannot be cleanly selected by thresholding without some processing.
      • The image has noise, which can be removed by smoothing.
      • The illumination (over-all signal level) falls off to the left.
    • The easiest way around the problem with this particular image is to divide it in half, and measure the left and right halves separately.
    • Alternatively, the image can be flattened by convolving a copy of the image several times with the Mean(63x63 ) kernel (this is the largest available in Image), and using the paste control to divide the original by the smoothed copy. (This does not appear to work properly with Scion Image - the gray levels are reduced such that the three phases are no longer distinguishable.)
    • FFT low cut filtering does not seem to work on this image as well as on the Betsy image in the FFT Flatten Illumination exercise. The requirement is more stringent here - the gray levels in each phase must be flat for thresholding, this is harder than making the image just look better. If too many low frequencies are removed , each phase becomes more shaded. If too few low frequencies are removed , the image still has a gray level gradient going across it. (The gradient is apparent if a red band, 1-2 cm long in the LUT window is moved up and down with the mouse, while observing the gray areas in the image.) Perhaps more precise low cut filtering, using the FFT macros, would be sufficient.