Parent Topic: CLWORKS

Training Site Editor

The ``Training Site Editing'' panel is invoked by selecting the similarly named option from the Classify menu item. It is also automatically invoked after the session configuration (with the Supervised option selected).

This panel is used for collecting training areas, assigning class attributes, and generating signatures. Used in conjunction with the training site are signature statistics, scatterplots, histograms, separability reports and a classification preview, all of which are available in the main Classify menu.

When an area is loaded, the training area channel is overlaid on the visible image in the main ImageWorks display window. All training areas collected during this or previous edits are displayed as an overlay using the colours of the class they represent. The visual channels and the training channel may be changed by going to the session configuration panel and changing the desired channels.

There are three sections to the training site editor.

The top area of the panel contains controls for the creation and modification of training sites.

Load: Pops up a panel to load the working area at 1:1 resolution.

Import Vectors: Use rasterized vectors to define training sites. See Import Vectors for more information.

Import Bitmaps: Uses bitmaps to define training sites. See Import Bitmaps for more information.

Trace: Trace a line, typically a border, for a future fill operation.

Trace & Close: Similar to Trace, except that the shape is automatically closed when the left mouse button is released.

Solid Box: Depress the left mouse button and drag out a box to define a solid (filled in) rectangular area:

Solid Circle: Depress the left mouse button and drag out a circle to define a solid (filled in) circular area.

Polygon: Pressing on the left mouse button will create several joined lined segments. Right clicking or selecting the polygon button again will close the figure. This figure may then be subsequently filled.

Polyline: Pressing on the left mouse button will create several joined lined segments.

Seed: This will fill in a region of similar pixels. This is a convenient aid when creating irregularly shaped training sites. See "Seed" for more information.

Fill: Pressing either the left or right mouse button over the image will cause a flood fill operation until a training area boundary is met. The right mouse button leaves Fill mode active.

Unfill: Similar to the fill operation, but used to delete a connected region of the training area for the current class.

Undo: Cancels the previous operation. To redo an operation, press undo again.

Erase: Puts you in erase mode. When used in conjunction with the size menu to the right, Erase creates an eraser of the indicated size. Pressing and moving the mouse in the main window will erase the underlying portions of the current class.

The second area of the panel contains a simple slider controlling the transparency of the overlaid training sites. The transparency may be varied from 0->100% ranging from opaque to completely invisible. When the transparency figure exceeds 50%, the indicator will flash red, warning the user that it may be difficult or even impossible to see the training site properly.

The last section of the panel contains a table listing the classes as well as several buttons. One of the rows in the table may be highlighted. This class is described as the current class. Most of the editing commands apply only to the current class.

The table has seven columns:

Value: The value field contains the ``encoding'' value for the class. This is the digital value placed in the classification output channel to mark areas that are part of this class. This value may be modified, but must be unique within the list. You can sort the list by value by clicking in the column header.

Name: This field is a short name for the class. Although there is no limit to its length, typically a short name is used. Click on the column header to sort by name. If this text appears red then there are classes in the list with duplicated names.

Colour: This colour chip indicates the colour used to represent this class. Clicking on the colour will pop-up a colour chooser, enabling you to modify the colour. If the header appears in red then there are classes using duplicate colours. Duplicate colours have an ``X'' in the colour box.

Threshold: This field contains a number measured in standard deviations. The value may be changed by entering a new value in the cell. The default value is 3.0 .

Bias: This field contains a number indicating the preference between classes when a pixel could be classified in more than one class. The class with the highest Bias value wins the ``tie''. The value may be changed by entering a new value in the cell. The default value is 1.0.

Manual: This indicates whether or not the signature of the class is to be automatically calculated from the training site, or fixed. A signature may be fixed by using an imported signature or by modifying the mean for a particular channel. This is fully described in the Signature Statistics help topic.

Description: Provides a longer description of the class. The list may be sorted by description by clicking on the header.

These fields may all be edited simply by clicking on the value that you wish to edit. The table cannot be edited if the background is gray, which typically means that a working area has not been loaded yet.

The table of class values and information on how to use it is completely described under the Class Lists topic.

Below the table are five buttons:

New: This button is used to create a new class (if permitted). You can create up to a maximum of 255 classes.

Merge: This button opens the ``Merge'' panel which is used to merge one or more different classes into another class.

Clear All: This will clear the training channel but will not delete the classes. This essentially sets the training channel all to zeros.

Delete: This will clear and delete all highlighted classes in the table.

Select Class at Cursor: Selecting this button will highlight the class at the current cursor locations. This may also be done by pressing the right mouse button in the main image window.

At the bottom are four action buttons:

Close: Pressing this button will save all edits and close the panel.

Preview: This button will perform a classification on the current working area. See the Preview Subtopic for more information.

Save: This button will save any edits made since the last save was made. Note that the training panel automatically saves all edits periodically.

Help: Displays help information.