Classification

Classifies the sample or samples located in the given region of interest (ROI) based on their shape. Refer to the NI Classification Training Interface Help for information about the NI Classification Training Interface controls.

Main Tab

The following controls are available on the Main tab.

• Step Name—Name to give the step.
• Reposition Region of Interest—When enabled, the step dynamically repositions the region of interest based on a coordinate system you built in a previous step.
• Reference Coordinate System—Coordinate system to which you want to link the region of interest.
• Samples—Indicates the number of samples learned in the selected classifier file.

Train Tab

The following controls are available on the Train tab.

• Classifier File Path—Path of the classifier file you want to use to classify samples. A classifier file contains a representation for each trained sample as well as its corresponding label.
• New Classifier File—Allows you to create a new classifier file using the NI Classification Training Interface.
• Sample to Classify—Specifies whether to classify only the largest sample in the region or all of the samples in the region.
• Training Required—Indicates that you modified some parameters that require the classifier to be trained before classifying samples.
• Samples—Indicates the number of samples learned in the selected classifier file.

Threshold Tab

The following controls are available on the Threshold tab.

• Method—Specifies whether to perform a manual or automatic threshold. To perform a manual threshold, select Manual Threshold from the Method drop-down menu. To perform an automatic threshold, select one of the following options:
  • Clustering—Sorts the histogram of the image within a discrete number of classes corresponding to the number of phases perceived in an image. Clustering is the most frequently used automatic thresholding method.
  • Entropy—Detects samples that are present in minuscule proportions on the image.
  • Metric—Calculates a value for each threshold that is determined by the surfaces representing the initial gray scale.
  • Moments—Use for images that have poor contrast.
  • Inter Variance—Use for images in which classes are not too disproportionate. For satisfactory results, the smallest class must be at least 5% of the largest one.
• Look For—Specifies whether you want to classify bright, dark, or gray objects.
• Range—Cluster specifying the threshold range.
  • Min—Lower value of the threshold range when you perform a manual threshold. In automatic threshold mode, Min displays the value computed by the automatic threshold algorithm you selected.
  • Max—Upper value of the threshold range when you use a manual thresholding method. In automatic threshold mode, Max displays the value computed by the automatic threshold algorithm you selected.
  • Lower Limit—Lower limit of the thresholding range when you use an automatic thresholding method. The automatic thresholding algorithm you select cannot compute a threshold value lower than Lower Limit.
  • Upper Limit—Upper Limit of the thresholding range when you use an automatic thresholding method. The automatic thresholding algorithm you select cannot compute a threshold value greater than Upper Limit.
• Ignore Objects Touching Region Borders—When enabled, ignores objects in the sample that are touching the border of the ROI.
• Remove Small Objects (# of Erosions)—Number of erosions you want the classification engine to perform to remove small objects in the sample from the ROI.
• Training Required—Indicates that you modified some parameters that require the classifier to be trained before classifying samples.
• Samples—Indicates the number of samples learned in the selected classifier file.

Options Tab

The following controls are available on the Options tab.

• Method—Method of classification.
  • Nearest Neighbor—Most direct approach to classification. In nearest neighbor classification, the distance of an input feature vector of unknown class to another class is defined as the distance to the closest samples that are used to represent that class.
  • K-Nearest Neighbor—More robust to noise compared with nearest neighbor classification. In K-nearest neighbor classification, an input feature vector is classified into a class based on a voting mechanism. The NI Classifier finds K nearest samples from all the classes. The input feature vector of unknown class is assigned to the class with majority of the votes in the K nearest samples.
  • Minimum Mean Distance—Most effective in applications that have little or no feature pattern variability or other corruptive influences. In minimum mean distance classification, an input feature vector of unknown class is classified based on its distance to each class center.
• Metric—Computes the distance between features in a classification application.
  • Maximum—Most sensitive to small variations between samples. Use Maximum when you need to classify samples with very small differences into different classes.
  • Sum—Metric used in most classification applications (also known as the Manhattan metric or Taxicab metric). This is the default value.
  • Euclidean—Least sensitive to small variations between samples. Use Euclidean when you need to classify samples with small differences into the same class.
• K—Sets the K value when using the K-Nearest Neighbor method of classification. The default is 3.
• Training Required—Indicates that you modified some parameters that require the classifier to be trained before classifying samples.
• Samples—Indicates the number of samples learned in the selected classifier file.

Parameters Tab

The following controls are available on the Parameters tab.

Parameters define the dependence of the classification engine on shape, scale, and mirror symmetry. By default, when the Scale Dependent and Mirror Dependent options are disabled, the NI Classifier depends only on variations in shape to classify samples. When Scale Dependent and Mirror Dependent are enabled, the dependence on shape is calculated as follows: Shape Dependence = 1000 – (Scale Factor + Mirror Factor).

• Scale Factor—Determines the relative importance (between 0 and 1000) of scale when classifying samples. If the value is 0, the samples are classified independent of scale.
• Mirror Factor—Determines the relative importance (between 0 and 1000) of mirror symmetry when classifying samples. If the value is 0, the samples are classified independent of mirror symmetry. Examples of objects exhibiting mirror symmetry are a lowercase letter p and a lowercase letter q.
• Training Required—Indicates that you modified some parameters that require the classifier to be trained before classifying samples.
• Samples—Indicates the number of samples learned in the selected classifier file.