Thresholding and Binarization

Use the Segmentation/ Threshold function to:

• binarize the image (set it black or white, binary)
• ceiling or bottom the image (discard values larger or smaller than a certain value)
• mask images directly, or use the binarized images for masking.

Binarized images consist of 32bit floating point pixels values, as any other image in Image Analyst MKII, but the actual pixel values are 1 or 0. In contrast image masks are using pixel value 'MASK' which is a non-numerical floating point value (e.g. result of 0/0).

The Threshold function offers the following methods of thresholding (see illustration below):

• For non-locally adaptive thresholding:
  • The Threshold from local max/min is None
  • The Threshold value calculation method is:
    • Pixel Value: a uniform threshold is applied to all images
    • Percentile by frames: the thresholding is adaptive, but uniform within each image. The level is calculated as the given percentile of the intensity histogram of each image.
    • Percentile by series: the thresholding is adaptive, but uniform within whole image series. The level is calculated as the given percentile of the intensity histogram of the image series.
    • Otsu by frames: the thresholding is adaptive, but uniform within each image. The level is calculated as the Otsu optimal threshold level for each image.
    • Otsu by series: the thresholding is adaptive, but uniform within whole image series. The level is calculated as the Otsu optimal threshold level for the complete image series.
• For locally adaptive thresholding:
  • The Threshold value calculation method is one of the above, but the calculated threshold value is used in a different way, based on the settings below:
  • The Threshold from local max/min is:
    • Bound Maxima uniformly: each object (part of the image) which is brighter than its surroundings by the above calculated threshold value, will be white (1). The surroundings will be black (0). Practically the threshold value is applied downwards from the local maximum. The Determine boundaries at is normally set to zero. This value is added the threshold value when determining the boundaries of an object, so increase this value to get thicker objects.
    • Bound Maxima locally: each object (part of the image) which is brighter than its surroundings by the above calculated threshold value, will be white (1). The surroundings will be black (0). The boundary of the object will be determined at the percent of its maximal intensity, given in the Determine boundaries at parameter.
    • Maximum Markers: only the local maxima are shown as white (1) pixels.
    • Bound Minima uniformly, Bound Minima locally, Minimum Markers: same as above but for minima. In this case set Way to Below.

Ways of thresholding:

• Above: sets pixels with greater or equal value than the (local) threshold to white (1), and the rest to black (0)
• Below: sets pixels with smaller or equal value than the (local) threshold to white (1), and the rest to black (0)
• Ceiling:  sets pixels with greater value than the threshold to the threshold value, the remaining pixels are unaffected
• Bottom:  sets pixels with smaller value than the threshold to the threshold value, the remaining pixels are unaffected
  Ceiling or Bottom used together with "Bound ... locally" will ceiling for maxima and bottom for minima at the local threshold. Ceiling and bottom are not available for "Bound ... uniformly" and for "Markers" and Above or Below binarization is performed instead.
• Mask Above: sets pixels with greater or equal value than the (local) threshold to MASK value
• Mask Below: sets pixels with smaller or equal value than the (local) threshold to MASK value

Examples are shown for adaptive and locally adaptive thresholding of an image with uneven background:

Image and parametering of Threshold	Cross-section intensities along the indicated ROI	Explanation (illustrations)
Test image (drawn in Corel PHOTO-PAINT and saved as 16bit TIF). Use Plot/Plot type=Line Scan to generate plots. The result of the binarization using different techniques is shown below:
Thresholding at an automatically determined level
Thresholding from local maxima brighter than the surroundings by 10000 units
Thresholding from local maxima maxima brighter than the surroundings by 30000 units
Thresholding from local maxima maxima brighter than the surroundings by 10000 units, calculating the threshold for each object at the 50% of peak intensity (as compared to the global minimum).