Package gov.nih.mipav.model.algorithms.levelset

Class LsePdeFilter2

java.lang.Object
gov.nih.mipav.model.algorithms.levelset.LsePdeFilter
gov.nih.mipav.model.algorithms.levelset.LsePdeFilter2
Direct Known Subclasses:
LseCurvatureFlow2, LseEvolve2, LseGaussianBlur2, LseGradientAnisotropic2
public abstract class LsePdeFilter2 extends LsePdeFilter

The abstract base class for finite-difference-based solvers for partial differential equations in 2D. This system exists to support level-set evolution for image segmentation.

The internal 2D data and mask images are copies of the inputs to the constructor but padded with a 1-pixel thick border to support filtering on the image boundary. These images are of size (xbound+2)-by-(ybound+2). The inputs (x,y) to access the data and mask avlues are constrained to 0 invalid input: '<'= x invalid input: '<' xbound and 0 invalid input: '<'= y invalid input: '<' ybound. The correct lookups into the padded arrays are handled internally.

Version:
0.1 November 7, 2006
Author:
David Eberly

  • Field Summary

    Fields
    Modifier and Type
    Field
    Description
    protected boolean[][]
    m_aabMask
    Pixels may be masked out so that the PDE solver does not process them.
    protected float[][]
    m_aafDst
    Successive iterations of the PDE solver toggle between two buffers.
    protected float[][]
    m_aafSrc
    Successive iterations of the PDE solver toggle between two buffers.
    protected float
    m_fFourthInvDxDy
    invFourthDxDy = 1/(4*dx*dy)
    protected float
    m_fHalfInvDx
    halfInvDx = 1/(2*dx)
    protected float
    m_fHalfInvDy
    halfInvDy = 1/(2*dy)
    protected float
    m_fInvDx
    invDx = 1/dx
    protected float
    m_fInvDxDx
    invDxDx = 1/(dx*dx)
    protected float
    m_fInvDy
    invDy = 1/dy
    protected float
    m_fInvDyDy
    invDyDy = 1/(dy*dy)
    protected float
    m_fUmm
    Temporary storage for 3x3 neighborhood.
    protected float
    m_fUmp
    Temporary storage for 3x3 neighborhood.
    protected float
    m_fUmz
    Temporary storage for 3x3 neighborhood.
    protected float
    m_fUpm
    Temporary storage for 3x3 neighborhood.
    protected float
    m_fUpp
    Temporary storage for 3x3 neighborhood.
    protected float
    m_fUpz
    Temporary storage for 3x3 neighborhood.
    protected float
    m_fUzm
    Temporary storage for 3x3 neighborhood.
    protected float
    m_fUzp
    Temporary storage for 3x3 neighborhood.
    protected float
    m_fUzz
    Temporary storage for 3x3 neighborhood.
    protected float
    m_fXSpacing
    The x-spacing of the image, call it dx.
    protected float
    m_fYSpacing
    The y-spacing of the image, call it dy.
    protected int
    m_iXBound
    The x-bound of the image.
    protected int
    m_iYBound
    The y-bound of the image.

    Fields inherited from class gov.nih.mipav.model.algorithms.levelset.LsePdeFilter

    m_eScaleType, m_fBorderValue, m_fMin, m_fOffset, m_fScale, m_fTimeStep, m_iQuantity, NONE, PRESERVE_ZERO, SYMMETRIC, UNIT

  • Constructor Summary

    Constructors
    Modifier
    Constructor
    Description
    protected
    LsePdeFilter2(int iXBound, int iYBound, float fXSpacing, float fYSpacing, float[] afData, boolean[] abMask, float fBorderValue, int eScaleType)
    Create a new PDE filter object for 2D images.

  • Method Summary

    Modifier and Type
    Method
    Description
    protected void
    assignDirichletImageBorder()
    Assign values to the 1-pixel-thick image border.
    protected void
    assignDirichletMaskBorder()
    Assign values to the 1-pixel-thick borders that surround unmasked regions.
    protected void
    assignNeumannImageBorder()
    Assign values to the 1-pixel-thick image border.
    protected void
    assignNeumannMaskBorder()
    Assign values to the 1-pixel-thick borders that surround unmasked regions.
    final boolean
    getMask(int iX, int iY)
    Get the mask value at pixel (x,y).
    final float
    getU(int iX, int iY)
    Get the image value at pixel (x,y), call it u(x,y).
    final float
    getUx(int iX, int iY)
    Get the first-order x-derivative of the image value at pixel (x,y), call it u_x(x,y).
    final float
    getUxx(int iX, int iY)
    Get the second-order x-derivative of the image value at pixel (x,y), call it u_xx(x,y).
    final float
    getUxy(int iX, int iY)
    Get the second-order mixed-derivative of the image value at pixel (x,y), call it u_xy(x,y).
    final float
    getUy(int iX, int iY)
    Get the first-order y-derivative of the image value at pixel (x,y), call it u_y(x,y).
    final float
    getUyy(int iX, int iY)
    Get the second-order y-derivative of the image value at pixel (x,y), call it u_yy(x,y).
    final int
    getXBound()
    Get the x-bound for the image.
    final float
    getXSpacing()
    Get the x-spacing for the image.
    final int
    getYBound()
    Get the y-bound for the image.
    final float
    getYSpacing()
    Get the y-spacing for the image.
    protected void
    lookUp5(int iX, int iY)
    Copy the source data to temporary storage in order to avoid redundant array accesses.
    protected void
    lookUp9(int iX, int iY)
    Copy the source data to temporary storage in order to avoid redundant array accesses.
    protected void
    onPostUpdate()
    Swap the buffers for the next pass of the PDE solver.
    protected void
    onPreUpdate()
    Recompute the boundary values when Neumann conditions are in effect.
    protected void
    onUpdate()
    Iterate over all the pixels and call onUpdate(x,y) for each pixel that is not masked out.
    protected abstract void
    onUpdate(int iX, int iY)
    An abstract function that allows per-pixel processing by the PDE solver.

    Methods inherited from class gov.nih.mipav.model.algorithms.levelset.LsePdeFilter

    getBorderValue, getQuantity, getScaleType, getTimeStep, setTimeStep, update

    Methods inherited from class java.lang.Object

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

  • Field Details

    • m_iXBound

      protected int m_iXBound
      The x-bound of the image.

    • m_iYBound

      protected int m_iYBound
      The y-bound of the image.

    • m_fXSpacing

      protected float m_fXSpacing
      The x-spacing of the image, call it dx.

    • m_fYSpacing

      protected float m_fYSpacing
      The y-spacing of the image, call it dy.

    • m_fInvDx

      protected float m_fInvDx
      invDx = 1/dx

    • m_fInvDy

      protected float m_fInvDy
      invDy = 1/dy

    • m_fHalfInvDx

      protected float m_fHalfInvDx
      halfInvDx = 1/(2*dx)

    • m_fHalfInvDy

      protected float m_fHalfInvDy
      halfInvDy = 1/(2*dy)

    • m_fInvDxDx

      protected float m_fInvDxDx
      invDxDx = 1/(dx*dx)

    • m_fFourthInvDxDy

      protected float m_fFourthInvDxDy
      invFourthDxDy = 1/(4*dx*dy)

    • m_fInvDyDy

      protected float m_fInvDyDy
      invDyDy = 1/(dy*dy)

    • m_fUmm

      protected float m_fUmm
      Temporary storage for 3x3 neighborhood. In the notation m_fUij, the i and j indices are in {m,z,p}, referring to subtract 1 (m), no change (z), or add 1 (p) to the appropriate index.

    • m_fUzm

      protected float m_fUzm
      Temporary storage for 3x3 neighborhood. In the notation m_fUij, the i and j indices are in {m,z,p}, referring to subtract 1 (m), no change (z), or add 1 (p) to the appropriate index.

    • m_fUpm

      protected float m_fUpm
      Temporary storage for 3x3 neighborhood. In the notation m_fUij, the i and j indices are in {m,z,p}, referring to subtract 1 (m), no change (z), or add 1 (p) to the appropriate index.

    • m_fUmz

      protected float m_fUmz
      Temporary storage for 3x3 neighborhood. In the notation m_fUij, the i and j indices are in {m,z,p}, referring to subtract 1 (m), no change (z), or add 1 (p) to the appropriate index.

    • m_fUzz

      protected float m_fUzz
      Temporary storage for 3x3 neighborhood. In the notation m_fUij, the i and j indices are in {m,z,p}, referring to subtract 1 (m), no change (z), or add 1 (p) to the appropriate index.

    • m_fUpz

      protected float m_fUpz
      Temporary storage for 3x3 neighborhood. In the notation m_fUij, the i and j indices are in {m,z,p}, referring to subtract 1 (m), no change (z), or add 1 (p) to the appropriate index.

    • m_fUmp

      protected float m_fUmp
      Temporary storage for 3x3 neighborhood. In the notation m_fUij, the i and j indices are in {m,z,p}, referring to subtract 1 (m), no change (z), or add 1 (p) to the appropriate index.

    • m_fUzp

      protected float m_fUzp
      Temporary storage for 3x3 neighborhood. In the notation m_fUij, the i and j indices are in {m,z,p}, referring to subtract 1 (m), no change (z), or add 1 (p) to the appropriate index.

    • m_fUpp

      protected float m_fUpp
      Temporary storage for 3x3 neighborhood. In the notation m_fUij, the i and j indices are in {m,z,p}, referring to subtract 1 (m), no change (z), or add 1 (p) to the appropriate index.

    • m_aafSrc

      protected float[][] m_aafSrc
      Successive iterations of the PDE solver toggle between two buffers. This is the source buffer, which is the input to the solver.

    • m_aafDst

      protected float[][] m_aafDst
      Successive iterations of the PDE solver toggle between two buffers. This is the destination buffer, which is the output of the solver.

    • m_aabMask

      protected boolean[][] m_aabMask
      Pixels may be masked out so that the PDE solver does not process them. This is useful for filtering only those pixels in a subimage.

  • Constructor Details

    • LsePdeFilter2

      protected LsePdeFilter2(int iXBound, int iYBound, float fXSpacing, float fYSpacing, float[] afData, boolean[] abMask, float fBorderValue, int eScaleType)
      Create a new PDE filter object for 2D images.
      Parameters:
      iXBound - The x-bound of the image.
      iYBound - The y-bound of the image.
      fXSpacing - The x-spacing of the image.
      fYSpacing - The y-spacing of the image.
      afData - The image elements, stored in lexicographical order.
      abMask - The image mask, stored in lexicographical order. A pixel value is processed by the PDE solver only when the mask value is false.
      fBorderValue - Specifies how to handle the image value. When set to Float.MAX_VALUE, Neumann conditions are in use, in which case zero-valued derivatives are assumed on the image border. Otherwise, Dirichlet conditions are used, in which case the image is assumed to be constant on the border with value specified by fBorderValue.
      eScaleType - The type of scaling to apply to the input image. The choices are NONE, UNIT, SYMMETRIC, or PRESERVE_ZERO.

  • Method Details

    • getXBound

      public final int getXBound()
      Get the x-bound for the image.
      Returns:
      The x-bound for the image.

    • getYBound

      public final int getYBound()
      Get the y-bound for the image.
      Returns:
      The y-bound for the image.

    • getXSpacing

      public final float getXSpacing()
      Get the x-spacing for the image.
      Returns:
      The x-spacing for the image.

    • getYSpacing

      public final float getYSpacing()
      Get the y-spacing for the image.
      Returns:
      The y-spacing for the image.

    • getU

      public final float getU(int iX, int iY)
      Get the image value at pixel (x,y), call it u(x,y).
      Parameters:
      iX - The x-coordinate of the pixel.
      iY - The y-coordinate of the pixel.
      Returns:
      The image value, u(x,y).

    • getUx

      public final float getUx(int iX, int iY)
      Get the first-order x-derivative of the image value at pixel (x,y), call it u_x(x,y).
      Parameters:
      iX - The x-coordinate of the pixel.
      iY - The y-coordinate of the pixel.
      Returns:
      The x-derivative of the image value, u_x(x,y).

    • getUy

      public final float getUy(int iX, int iY)
      Get the first-order y-derivative of the image value at pixel (x,y), call it u_y(x,y).
      Parameters:
      iX - The x-coordinate of the pixel.
      iY - The y-coordinate of the pixel.
      Returns:
      The y-derivative of the image value, u_y(x,y).

    • getUxx

      public final float getUxx(int iX, int iY)
      Get the second-order x-derivative of the image value at pixel (x,y), call it u_xx(x,y).
      Parameters:
      iX - The x-coordinate of the pixel.
      iY - The y-coordinate of the pixel.
      Returns:
      The xx-derivative of the image value, u_xx(x,y).

    • getUxy

      public final float getUxy(int iX, int iY)
      Get the second-order mixed-derivative of the image value at pixel (x,y), call it u_xy(x,y).
      Parameters:
      iX - The x-coordinate of the pixel.
      iY - The y-coordinate of the pixel.
      Returns:
      The xy-derivative of the image value, u_xy(x,y).

    • getUyy

      public final float getUyy(int iX, int iY)
      Get the second-order y-derivative of the image value at pixel (x,y), call it u_yy(x,y).
      Parameters:
      iX - The x-coordinate of the pixel.
      iY - The y-coordinate of the pixel.
      Returns:
      The yy-derivative of the image value, u_xx(x,y).

    • getMask

      public final boolean getMask(int iX, int iY)
      Get the mask value at pixel (x,y).
      Parameters:
      iX - The x-coordinate of the pixel.
      iY - The y-coordinate of the pixel.
      Returns:
      The mask value at pixel (x,y). Its value is true when that pixel should not be processed by the PDE solver.

    • assignDirichletImageBorder

      protected void assignDirichletImageBorder()
      Assign values to the 1-pixel-thick image border. The border value has been specified during object construction.

    • assignNeumannImageBorder

      protected void assignNeumannImageBorder()
      Assign values to the 1-pixel-thick image border. The border values are chosen so that the first-order image derivatives at the border pixels are zero. Thus, the border values are duplicates of the adjacent image values.

    • assignDirichletMaskBorder

      protected void assignDirichletMaskBorder()
      Assign values to the 1-pixel-thick borders that surround unmasked regions. The border value has been specified during object construction.

    • assignNeumannMaskBorder

      protected void assignNeumannMaskBorder()
      Assign values to the 1-pixel-thick borders that surround unmasked regions. The border values are chosen so that the first-order image derivatives at the border pixels are zero. Thus, the border values are duplicates of the adjacent image values.

    • onPreUpdate

      protected void onPreUpdate()
      Recompute the boundary values when Neumann conditions are in effect. If a derived class overrides this, it must call the base-class onPreUpdate first.
      Specified by:
      onPreUpdate in class LsePdeFilter

    • onUpdate

      protected void onUpdate()
      Iterate over all the pixels and call onUpdate(x,y) for each pixel that is not masked out.
      Specified by:
      onUpdate in class LsePdeFilter

    • onPostUpdate

      protected void onPostUpdate()
      Swap the buffers for the next pass of the PDE solver. If a derived class overrides this, it must call the base-class onPostUpdate last.
      Specified by:
      onPostUpdate in class LsePdeFilter

    • onUpdate

      protected abstract void onUpdate(int iX, int iY)
      An abstract function that allows per-pixel processing by the PDE solver. The pixel (x,y) must be in padded coordinates, namely, 1 invalid input: '<'= x invalid input: '<'= xbound and 1 invalid input: '<'= y invalid input: '<'= ybound.
      Parameters:
      iX - The x-coordinate of the pixel.
      iY - The y-coordinate of the pixel.

    • lookUp5

      protected void lookUp5(int iX, int iY)
      Copy the source data to temporary storage in order to avoid redundant array accesses. The copy involves the 5 pixels (x,y), (x+1,y), (x-1,y), (x,y+1), and (x,y-1).

    • lookUp9

      protected void lookUp9(int iX, int iY)
      Copy the source data to temporary storage in order to avoid redundant array accesses. The copy involves the 9 pixels in the 3x3 neighborhood of (x,y).