I have made some changes to make itk::PathToImageFilter to make it a usable base class. There were 
some problems with the CVS version, including: (1) this ase classhad actual filter functionality, (2) 
the mechanism for specifying the size, spacing, origin, etc., of the output image was incomplete, contrary 
to ITK convention, and hard to extend, and (3) the documentation specified that the filter would calculate 
the size of the output image from the path, but it did not. 

I solved these problems by (1) moving the filter functionality to itk::TracePathOnImageFilter, (2) 
requiring that filter->GetOutput()->Set...(...) should be used to set the information for the 
output image, and (3) providing methods to calculate the correct image region a path inhabits, and a 
boolean flag to specify whether that calculation should be done (default true).

Changes to itk::PathToImageFilter.

The report details some modifications made to the ITK
{\em ConnectedComponentImageFilter} in an attempt to improve
performance. Some interesting observations were made during this
process. A new filter using a different algorithm to perform the same
function is also described and improved performance demonstrated.

Optimization of connected component labelling

Morphological reconstruction may be implemented in a number of
different ways. ITK has an iterative method and a non iterative
method. This article compares the performance of another non iterative
method and finds a significant improvement.

Improving performance of morphological reconstruction

Fast computation of distance transforms find direct application
in various computer vision problems. Currently there exists 
two image filters in the ITK library which can be used to generate
distance maps. Unfortunately, these 
filters produce only approximations to the Euclidean Distance
Transform (EDT). We introduce into the ITK library a third EDT 
filter which was developed by Maurer {\em et al.} \cite{Maurer2003}.
In contrast to other algorithms, this algorithm produces the 
exact signed squared EDT using integer arithmetic. The complexity,
which is formally verified, is $O(n)$ with a small time constant
where $n$ is the number of image pixels.

N-D Linear Time Exact Signed Euclidean Distance Transform

The itk::GaussianDerivativeImageFunction computes the derivative at the specified physical or pixel location. Unfortunately it has a number of deficiencies, for which I suggest possible solutions.

A small rework for the Gaussian Derivative Image Function

LabelOverlayImageFilter colorize a label image and put it on top of the input image.

Label overlay

Binary erosion and dilation are the base filters of binary mathematical morphology.
In ITK 2.4.1 \verb$BinaryDilateImageFilter$ and \verb$BinaryErodeImageFilter$ implement
an efficient algorithm described in \cite{Nikopoulos2000}. However, those filter lack support for boundary values, have a poor progress report, and can be quite inefficient
for complex 3D images.

Improving features and performance of binary erode and dilate filters

This document describes the iteration direction optimized Image Iterators . Comparing with the original corresponding iterators, the new iterators provide automatic iteration direction optimization and give more flexible. They are more efficient when the fastest moving direction has small size. The derived template class ImagePieceWiseLinearIteratorWithIndex is a generic class that can be replace the ImageLinearIterator and ImageSliceIterator. The paper present the implementation and examples of these classes. This paper is accompanied with the source code and experiments.

optimized image iterators

ITK is a huge image analysis library, which contains lots of state of the arts
algorithms implementations. However, using it in C++ can be difficult and is
definitively bad suited for prototyping. WrapITK aims to allow classes from ITK
(and custom, classes that interact with ITK) to be "wrapped" for use with
languages like Python, Tcl, and Java.

WrapITK: Enhanced languages support for the Insight Toolkit

This document describes contributions on the
NormalizedCorrelationImageToImageMetric and the MeanSquaresImageToImageMetric of the Insight Toolkit ITK \url{www.itk.org}. For the first metric a two time speed-up can be
achieved by rewriting the code to loop only once over the fixed image. This is instead of the two times that is used in the current ITK code. The reduction in computation time comes at the cost of an additional storage of a parameters array. For both metric we have
implemented the option to use only a random subset of the fixed image voxels for calculating the metric value and its derivatives. This reduces the computation time (substantially), while convergence
properties are maintained. This paper is accompanied with the source code.