NearOCR

Class Symbol

java.lang.Object

NearOCR.CharacterImage

NearOCR.Symbol

public class Symbol
extends CharacterImage

A symbol is responsible for managing training and finding values from net output. It comprises a very important part of the post parsing system. Each symbol stores its character value (may be multiple characters, but is unlikely to be), the image from which the net was trained (used for on-the-fly adjustments to the nets, a reference to the original image location (used during save/load operations - symbols do not store images within their file formats) and an array of all weighted edges, the length of which must match the number of output neurons for the whole OCR manager.

Version:

0.1

Author:

Stephen Wattam

Field Summary

Fields inherited from class NearOCR.CharacterImage

image, value

Constructor Summary

Constructors

Constructor and Description
Symbol(java.lang.String p_value, java.lang.String p_imageFilename) Creates a new symbol with a given value and from a given filename.

Method Summary

Methods

Modifier and Type	Method and Description
double[]	getDesiredValues() This returns a double array of all 'desirable' edges which should be firing in order to trigger the symbol.
double[]	getDesiredValues(int p_start, int p_length) Returns a range of desired values.
int	getLength() Returns the number of weighted nodes held by this symbol.
double	getMeanDeviation(double[] p_outputs) Calculates a confidence value for this nsymbol based on net outputs.
WeightedNode	getWeight(int p_index) Returns one weighted node, addressed by the index in the internal array - NOT the index stored within the weighted node!
WeightedNode[]	getWeights() Returns a list of all weighted nodes stored within this Symbol.
WeightedNode[]	getWeightsFromOffset(int p_offset, int p_length) Returns a range of weighted nodes from a given position.
void	setLength(int p_length) Sets the number of output nodes, which ought to be equal to the number of total output nodes in the OCR manager.
void	setWeight(WeightedNode p_weight, int p_index) Sets the weight of one particular output node, indexed by value in internal array - NOT by the index as stored in the weighted node itself.
java.lang.String	toString() Returns this Symbol as represented by a string.

Methods inherited from class NearOCR.CharacterImage

getImage, getValue, setImage, setValue

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

Symbol

public Symbol(java.lang.String p_value,
      java.lang.String p_imageFilename)
       throws java.io.IOException

Creates a new symbol with a given value and from a given filename.

Parameters:

p_value - The initial value of this symbol

p_imageFilename - A path to the image to load. The constructor loads the image itself

Throws:

java.io.IOException - in the event that any problem occurs when loading the image filepath

Method Detail

getWeights

public WeightedNode[] getWeights()

Returns a list of all weighted nodes stored within this Symbol. Its length should be equal to the number of output neurons in the entire OCR manager, but this is dependent upon when it was last updated using setLength. There is no guarantee that getWeights().length == OCRManager.countOutputs().

Returns:

All weighted nodes, in order

See Also:

setLength(int)

getWeightsFromOffset

public WeightedNode[] getWeightsFromOffset(int p_offset,
                                  int p_length)

Returns a range of weighted nodes from a given position.

Parameters:

p_offset - The start position

p_length - The number of items to return

Returns:

An array of WeightedNodes with length of p_length

getDesiredValues

public double[] getDesiredValues(int p_start,
                        int p_length)

Returns a range of desired values.

Parameters:

p_start - The start position.

p_length - The number of items to return

Returns:

A list of all desired values requested in the form of an array of doubles

getWeight

public WeightedNode getWeight(int p_index)

Returns one weighted node, addressed by the index in the internal array - NOT the index stored within the weighted node!

Parameters:

p_index - The index number to return the node from

Returns:

The weighted node at the index given

setWeight

public void setWeight(WeightedNode p_weight,
             int p_index)

Sets the weight of one particular output node, indexed by value in internal array - NOT by the index as stored in the weighted node itself.

Parameters:

p_weight - The weight to set

p_index - The index into which to insert this weighted node.

getMeanDeviation

public double getMeanDeviation(double[] p_outputs)

Calculates a confidence value for this nsymbol based on net outputs.

Parameters:

p_outputs - An array of all net outputs from this OCRManager.

Returns:

The mean confidence for all outputs

toString

public java.lang.String toString()

Returns this Symbol as represented by a string. USed in the save/load process

Overrides:

toString in class java.lang.Object

Returns:

This object, as represented in a string form.

setLength

public void setLength(int p_length)

Sets the number of output nodes, which ought to be equal to the number of total output nodes in the OCR manager. This scales an array and hence is probably very slow. Sizes which are smaller than current values will crop weights, those above will stuff the array with default values.

Parameters:

p_length - The number of nodes

getLength

public int getLength()

Returns the number of weighted nodes held by this symbol.

Returns:

The number of weighted nodex held by this symbol.

getDesiredValues

public double[] getDesiredValues()

This returns a double array of all 'desirable' edges which should be firing in order to trigger the symbol. It is one of the few methods that uses the indices stored within the nodes themselves.

Returns:

The desired values of all nodes.