Thresholds tab
The Thresholds tab contains the following groups, each with their own options:
-
Classification
-
Trainingset Extension
-
Filter Options
The Classification group contains the following options:
- Threshold for classification
-
Set the classification threshold by using the Threshold for classification slider or entering a number between 0 and 100%.
The classification threshold defines the similarity that an image has to have to be added to an existing class. The default setting is 70%. Below 60%, classification errors are likely to occur. Above 70%, more classes than necessary are created.
The Trainingset Extension group contains the following options:
- Threshold for trainingset extension
-
Set the threshold for training set extension by using the Threshold for training set extension slider or entering a number between 0 and 100%.
The training set of a class is automatically extended during clustering by good examples that have some features that are not already learned in the pattern of a class. The threshold for training set extension defines the minimum amount of similarity that an image has to have to be used as a new sample image in the training set. Increasing this value adds fewer samples to the training set. The default setting is 95%. If this threshold is below 80%, classification errors are likely to occur.
The Filter Options group contains the following options:
- Move dark and white images into the class "Null"
-
Enable this option if the document contains any unusable, blurred, dark, or white images that do not contain enough information to be classified. These images can be assigned to a null class. This ensures that these documents are not included in the training set or the classifier.
This option is cleared by default.
- Remove folders containing only one image after a certain number of steps and move the image into the class "NoMatch"
-
Enable this option to eliminate any class folders that contain a single document. It is likely that this document is not correctly classified, so moving it would improve subsequent results using that training set.
This option is cleared by default.
- Number of steps
-
If the Remove folders option is selected, this option is enabled. You can specify how many steps should run before any class folders containing a single document are moved to the "NoMatch" class. Set this number by using the Number of steps slider or enter a number between 0 and 1000.
This option is set to 1000 by default.