Facemorph.psm
Class PSMNode

java.lang.Object
  Facemorph.psm.PSMNode

public class PSMNode
extends java.lang.Object

PSMNode: an implementation of Pictorial Structure Models

Constructor Summary

PSMNode()
PSMNode constructor

PSMNode(int start, BigMat mst, Feature appGauss, DiagonalGaussian[][] gauss, boolean[] visited, int w, int h)
Contruct a PSMNode

Method Summary

void	batchTest(java.lang.String outfold, java.util.ArrayList<java.lang.String> imageNames, java.util.ArrayList<java.lang.String> templateNames, int w, HaarReader haar) Batch testing method
FloatImage	evaluateProbability(java.awt.Image img) Evaluate the probability of this feature in the image supplied
static int	findBestRoot(BigMat tree, BigMat graph) Finds the "best" root of the tree
static Feature	getFeature(java.lang.String className) Dynamically loads the Feature given its class name
java.util.ArrayList<java.awt.geom.Point2D.Float>	getFeatures() Get the feature points
Template	getFeatures(java.awt.image.BufferedImage bimg, HaarReader haar) Detect the deature in the image using this model
java.util.ArrayList<java.awt.geom.Point2D.Float>	getFeatures(java.awt.Image source) Get the feature points in the image
static void	main(java.lang.String[] args) Main method for testing
static BigMat	minimumSpanningTree(BigMat graph) Get the minimum spanning tree of the graph
static double	ncc(double[] v1, double[] v2) The normalised cross correlation
void	read(java.io.InputStream in, java.lang.String folder, boolean url) Read from the named input stream
void	read(java.io.StreamTokenizer st, Feature app, java.lang.String folder, boolean url) Read a PSM
void	read(java.lang.String fileName) Read from a file
void	read(java.lang.String fileName, boolean url) Read a PSMNode from file
double	shapeError(java.util.ArrayList<java.awt.geom.Point2D.Float> detectedPoints, Template template, int[] pointList) Find the RMS error
static PSMNode	train(java.util.ArrayList<java.lang.String> imageNames, java.util.ArrayList<java.lang.String> templateNames, int[] pointList, int w, int enrich, java.lang.String featureClassName, HaarReader haar) Train a PSM model
void	write(java.io.PrintStream ps, boolean writeApp, java.lang.String folder) Write a PSMNode
void	write(java.lang.String fileName) Write a PSMNode to the file named

Methods inherited from class java.lang.Object

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

Constructor Detail

PSMNode

public PSMNode()

PSMNode constructor

PSMNode

public PSMNode(int start,
               BigMat mst,
               Feature appGauss,
               DiagonalGaussian[][] gauss,
               boolean[] visited,
               int w,
               int h)

Contruct a PSMNode

Parameters:

start - the node to start at

mst - the minimum spanning tree

appGauss - the appearance

gauss - the conditional positional probability model of the nodes

visited - indicates which nodes have been visited

w - the width

h - the height

Method Detail

write

public void write(java.lang.String fileName)

Write a PSMNode to the file named

Parameters:

fileName - the name of the file

write

public void write(java.io.PrintStream ps,
                  boolean writeApp,
                  java.lang.String folder)

Write a PSMNode

Parameters:

ps - the print stream to write to

writeApp - Should the appearance be written

folder - the folder to write to

read

public void read(java.lang.String fileName,
                 boolean url)

Read a PSMNode from file

Parameters:

fileName - the name of the file or URL

url - if true the name referes to a URL

read

public void read(java.lang.String fileName)

Read from a file

Parameters:

fileName - the name of the file

read

public void read(java.io.InputStream in,
                 java.lang.String folder,
                 boolean url)
          throws java.io.IOException

Read from the named input stream

Parameters:

in - the input stream

folder - folder to read (children) from

url - indicates that the children should be read from a URL

Throws:

java.io.IOException

getFeature

public static Feature getFeature(java.lang.String className)

Dynamically loads the Feature given its class name

Parameters:

className - Name of the class to load, must implement the Feature interface and be in the class path to be loaded

Returns:

returns the Feature of the given name

read

public void read(java.io.StreamTokenizer st,
                 Feature app,
                 java.lang.String folder,
                 boolean url)
          throws java.io.IOException

Read a PSM

Parameters:

st - the stream tokenizer to read from

app - the appearance of this node

folder - indicates the base folder/url to read children from

url - indicates children should be read from a URL

Throws:

java.io.IOException

train

public static PSMNode train(java.util.ArrayList<java.lang.String> imageNames,
                            java.util.ArrayList<java.lang.String> templateNames,
                            int[] pointList,
                            int w,
                            int enrich,
                            java.lang.String featureClassName,
                            HaarReader haar)

Train a PSM model

Parameters:

imageNames - the list of training images

templateNames - the list of training template names

pointList - the list of points to use from the template

w - the desired width for training the appearance model

enrich - the amount of enrichment of the training data to use (currently disabled)

featureClassName - the name of the feature model to use

haar - the face detector to use

Returns:

returns the trained model

findBestRoot

public static int findBestRoot(BigMat tree,
                               BigMat graph)

Finds the "best" root of the tree

Parameters:

tree - contains the MST, with 1 or 0 to indicate tree edges

graph - the graph contains the edge weights of the complete graph

Returns:

returns the index of the most conected / central node

minimumSpanningTree

public static BigMat minimumSpanningTree(BigMat graph)

Get the minimum spanning tree of the graph

Parameters:

graph - the edge weight matrix

Returns:

return the MST as a matrix

evaluateProbability

public FloatImage evaluateProbability(java.awt.Image img)

Evaluate the probability of this feature in the image supplied

Parameters:

img - the image to test, or if null the currently set image is used

Returns:

returns the probability of the feature location, recursing down and back up the feature tree

getFeatures

public java.util.ArrayList<java.awt.geom.Point2D.Float> getFeatures(java.awt.Image source)

Get the feature points in the image

Parameters:

source - the input image

Returns:

the estimated feature locations

getFeatures

public java.util.ArrayList<java.awt.geom.Point2D.Float> getFeatures()

Get the feature points

Returns:

return the most likely feature points

ncc

public static double ncc(double[] v1,
                         double[] v2)

The normalised cross correlation

Parameters:

v1 - the first vector

v2 - the second vector

Returns:

returns the ncc between the two vectors

shapeError

public double shapeError(java.util.ArrayList<java.awt.geom.Point2D.Float> detectedPoints,
                         Template template,
                         int[] pointList)

Find the RMS error

Parameters:

detectedPoints - the detected points

template - the ground truth

pointList - the list of points in template that correspond to the detected points

Returns:

return the RMS error

batchTest

public void batchTest(java.lang.String outfold,
                      java.util.ArrayList<java.lang.String> imageNames,
                      java.util.ArrayList<java.lang.String> templateNames,
                      int w,
                      HaarReader haar)

Batch testing method

Parameters:

outfold - the output folder

imageNames - the names of the images to test

templateNames - the names of the templates

w - the desired appearance image size

haar - the face detector

getFeatures

public Template getFeatures(java.awt.image.BufferedImage bimg,
                            HaarReader haar)

Detect the deature in the image using this model

Parameters:

bimg - the image to search

haar - the face detector

Returns:

returns a Template containin the detected features, or null if the face detection failed

main

public static void main(java.lang.String[] args)

Main method for testing

Parameters:

args - the program args