Overview of Using CNLSearch

The process of using CNLSearch can be divided into the training-time steps and the search-time steps.

In general, you should follow these training-time steps:

Choose between linear and nonlinear mode search. If you are not sure which mode your application requires, you can train a model for both modes.
If you are using linear mode search, select your algorithm. You can train the model for both linear mode algorithms.
Select the search accuracy level. You can train the model for any or all accuracy levels.
Select a model image and train the model.

In general, you should follow these search-time steps:

Perform a series of test searches at different accuracy levels and with different algorithms.
Fine-tune the various search parameters, including the acceptance and confusion thresholds using the results of the test searches.
(Optional) Re-train the model using just the algorithms and accuracies that work best for your application. This can reduce the amount of memory required to store the model, but it does not change the speed of subsequent searches.

The table below contains an overview of the different training-time and run-time parameters you supply to CNLSearch.

	Parameter	Values	Notes
Training	Accuracy	Coarse Fine Veryfine	You must train the model for all accuracy levels you intend to search with. Training for additional accuracy levels increases training time and the amount of memory required for the model.
	Algorithm	CnlpasLinear CnlpasNonlinear Search	You must train the model for all the algorithms you intend to search with. Training for additional algorithms increases training time and the amount of memory required for the model.
	Algorithm	Normalized Absolute	Training for both normalized and absolute scoring has no cost at training time
Search	Accuracy	Coarse Fine Veryfine	Specify the search accuracy. The greater the accuracy, the slower the search. For more information, see the section Selecting a Search Accuracy.
	Algorithm	CnlpasLinear CnlpasNonlinear Search	Use the guidelines in the section Choosing Between Linear and Nonlinear Mode.
	Algorithm	Normalized Absolute	Use the guidelines in the section Choosing Between Normalized and Absolute Scoring.
	Confusion Threshold	0.0 - 1.0	The best score that a non-instance of the model can receive. For more information, see the section Selecting a Confusion Threshold and an Acceptance Threshold.
	Acceptance Threshold	0.0 - 1.0	The worst score that an instance of the model can receive. For more information, see the section Selecting a Confusion Threshold and an Acceptance Threshold

CNLSearch parameter overview