Quick Segmentation

Quick segmentation Uses a highly efficient segmentation algorithm to separate features, converting an image into a line drawing in which the lines depict outlines of objects in the image. Its purpose is to reduce the complexity of the image for pattern counting or further analysis. The algorithm is rapid but does not always give perfect segmentation.

The image to be segmented should be converted to grayscale and 8 bits/pixel for best results. If the segmentation appears incorrect, try one or more of the following:

• Select ``Color..Colorgrayscale'' to make sure tnimage is handling the image as grayscale. (Image may appear to be shades of gray and still be some other colortype).
• Invert the colors and segment the negative of the image. This often works if areas in the background were outlined.
• Try changing Maximum Signal from Black to White.
• Try changing Structuring element from N8 to N4.
• Change the brightness or contrast of the image.
• Filter the image to flatten the background using the following parameters:
  • Filter type: Remove low frequencies
  • Kernel: 15$\times$15
  • Kernel multiplier: 2
  • Amount of filtering: 100

The structuring element is the morphological kernel that determines how pixels are added or removed. Two structuring elements can be selected in tnimage:

• N4 consists of the 4 closest pixels, directly above, below, left, and right of each pixel.
• N8 consists of all 8 of the nearest neighbors to the pixel. N8 generally produces a smoother result, but takes longer.

This parameter is not used by watershed segmentation, which uses the ``Watershed kernel size'' parameter instead.

Quick segmentation of an image of a 2D protein gel.
A Original image
B Quick segmentation classifies much of upper left of image as a big feature.
C Same image filtered to remove low frequencies (kernel = 15x15, multiplier = 1).
D Quick segmentation of filtered image (threshold = 180).

Binary vs Graylevel - Binary filtering produces a solid white or black result, while graylevel filtering retains the original shades of gray. This is also not used by segmentation.

Maximum signal - If the features of interest in an image are black, this should be set to black. For grayscale erosion, switching the maximum signal is the same as switching from erosion to dilation.

Threshold - Determines the cutoff value for deciding whether a pixel is a feature or the background. A higher threshold causes more pixels to be considered features and produces more lines in the segmentation.

Watershed kernel - Used for watershed segmentation instead of structuring element. Determines distance from pixel to be considered as part of a neighborhood group. Large values will take a long time.

Notes

1. For quick segmentation, a threshold that is too high or too low will produce an all-white image.
2. For watershed segmentation, excessive low-pass smoothing can result in no watersheds being found, while no smoothing can result in over-segmentation.

Contour Map Uses the quick segmentation algorithm to create a new image consisting of contours following areas of steepest gradient. The image should be low-pass filtered first to avoid drawing circles around small noise features. The number of contours is determined by the ``Contour sep'' setting. A larger value results in fewer lines but is faster. The actual number of contours also depends on the image. The product is still an image and does not contain vectorized data. Note that it is not a true iso-intensity contour map since the number of lines is determined by the gradient.

Contour map Image was low-pass filtered with a 9$\times$9 kernel before creating the map using a spacing of 10. Left: Original, Center:Contours, Right: Contours superimposed on original image.