Image Segmentation

An image that has 256 values needs to be processed in such a way as to allow quantification by reducing the available gray values in an image to only the features of interest. The process in which 256 gray values reduced to 2 gray values is called thresholding.

  • Histogram segmentation
  • Non-uniform segmentation
  • Object morphology and Contiguity
  • Texture thresholding
  • Combinations
  • Pattern matching
  • Color segmentation

Histogram Segmentation

Interactive selection and Histogram Characteristics peaks thresholding methods are used frequently, sometimes together, depending on particular type of image being viewed.  Automatic thresholding often uses the histogram peaks method to determine where to set the grey level ranges for image segmentation.

One Level Thresholding

Original Grey Level Image

Segmented Image

Grey Level Historgram

Multi-level Thresholding

Original Grey Level Image

Grey Level Histogram

Segmented Image

Non-Uniform Segmentation

Original Grey Level Image

Delineation Filter

Binary Image

Object Morphology Example

Original Grey Level Image

Segmented Binary

features with length < 30mm - red  

features with sphericity < .65 - green

Object Contiguity Example

Original Grey Level Image

Two-Phase Segmented Binary Image

Red Phase Touchine Blue Phase

Texture Thresholding

Feature discrimination based on "texture' or "appearance', independent of size or color. Can be very challenging but processing algorithms getting very good!

Pattern Matching Example

Pattern-matching algorithms are powerful processing tools used to discriminate features of interest in an image. Usually, they require prior knowledge of the general shape of the   features contained in the image. For instance, if there are cylindrical fibers orientated in various ways within two-dimensional section of a composite, a set of boundaries can be generated that correspond to the angles at which a cylinder might occur in three-dimensional space.

Pattern-matching algorithms are used when required measurements cannot be directly made or calculated from the shape of a binary feature of interest.

Glass Fibers Fitted with Ellipses

                              Original Image

                               After template matching