Woolz is a set of software libraries and executables for image processing and pattern recognition that was initially developed at the MRC Clinical Population and Cytogenetics Unit. This became the MRC Human Genetics Unit (HGU) and in 2011 became part of the University of Edinburgh. Woolz was initially developed for fast microscope slide scanning, chromosome image analysis, pattern recognition and a wide range of image processing and analysis problems. The original authors of the software are Dr. Denis Rutovitz and Dr Jim Piper although the software has developed and expanded considerably since their initial work.
Woolz was adopted as the standard for the Mouse Atlas Databases (http://www.emouseatlas.org) and was used for all the reconstructions, anatomical, gene-expression and spatial domains. This was because the interval coding used in Woolz provided significant computing advantages in a range of image processing functions specifically in set operations (such as union, intersection, etc.) morphological operations (erosion, dilation etc.) and other binary image processing such as distance transforms, segmentation and labelling. In general Woolz is very efficient with respect to both memory use and time.
The Woolz data structures are, in general, compact and in terms of grey-level data minimise memory usage without compression (which in principle could also be applied). It is especially efficient for morphological and set operations because of the way it's 2 and 3D domain objects are encoded using intervals.
All pixel or voxel based image processing systems which allow image query via floating point coordinates need to define what a pixel is by the relationship between the integer pixel/voxel coordinates and the floating point query coordinates. In Woolz a pixel or voxel is defined to be an image sample centred at the integer coordinate. Following from this definition the C macro WLZ_NINT(), which returns the nearest integer to the given float, should be used in queries of pixel/voxel values when using floating point coordinates.
During it's use by the Mouse Atlas Project were many significant developments in Woolz, primarily focused on 3D reconstruction, transforms, registration and warping.
Since 2019 the development of Woolz has primarily been driven by the Gut Cell Atlas Project funded by the Helmsley Charitable Trust with significant developments including splines (for midline representation), JSON encoding and greater use from Python.
Woolz has been written in ANSI standard C so that it will build and run on all computing platforms that support some basic requirements, such as supporting at least 32 bit integers and IEEE floating point. The software is know to build on GNU/Linux systems, but will also build on MacOS or using MingW on Windows systems.
The code is partitioned into the following modules:
| Module | Description |
|---|---|
| libAlc | Library providing generic data structures and memory allocation functions |
| libAlg | Library providing basic numerical algorithms |
| libbibfile | Library with bibfile style input/output functions |
| libhguDlpList | Library with a generic doubly linked pointer list |
| libReconstruct | Library with code for 3D alignment of 2D section images to form a 3D image |
| libWlz | The Woolz image processing library |
| libWlzBnd | Library with small functions that bind Woolz to other languages |
| libWlzExtFF | Library for external data format input/output |
| binWlz | Small command line based Woolz programs |
| binWlzApp | More small command line based Woolz programs |
| binWlzExtFF | Small command line based Woolz programs which use external file formats |
| binWlzTst | Small command line based test programs for Woolz |
The authors include (in sort order):
| Bill Hill |
| Christophe Dubreuil |
| Elizabeth Guest |
| Jianguo Rao |
| Jim Piper |
| Konstantinos Liakos |
| Margaret Stark |
| Nick Burton |
| Richard Baldock |
To contact the authors please raise a Github issue.
Woolz is available as source from GitHub (https://github.com/ma-tech/) and can be downloaded pre-built for some architectures.
Woolz should build easily on most modern systems that have the GNU Build system (see http://en.wikipedia.org/wiki/GNU_build_system).
In most cases the simple script build.sh should be sufficient to build Woolz. This script simply runs autoreconf followed by ./configure with various options. It's probably best to copy build.sh to mybuild.sh and then set the required options for the build you want.
Sometimes the configuration will fail with messages relating to libtool. In these cases running: autoreconf -i --force may fix the problems.
Complains from configure that m4 can't be found can probably be fixed by running: automake --add-missing
The build script build.sh simply runs autoreconf followed by ./configure with the appropraite options.
A prefix can be given in the configure stage to define where the programs, libraries are will be installed, conventionally this is /opt/MouseAtlas. Use
./configure --help
to see all the options available.
There are over 200 small command line programs within Woolz. All of these accept -h as an argument to show their usage. On unix-like systems it is common to combine these small programs into a single command line with pipes. These programs include those for: applying affine, basis function and mesh based transforms registration of spatial domain objects and surfaces reconstruction from serial sections morphological operations (erosion, dilation, etc) and fill set operations (union, intersection etc) feature extraction mesh generation contour generation convex hulls distance transforms histograms thresholding and labelling grey and colour image value filters
As an example, the following thresholds a 3D image, applies erosion and dilation to remove small isolated regions, labels (segments) the image into separate objects and the prints the volume of each isolated object:
prompt% WlzThreshold -v135 -L ts14.wlz | \
WlzErosion -c26 -r2 | \
WlzDilation -c26 -r2 | \
WlzLabel | \
WlzVolume
Object 1: number of voxels = 66
Object 2: number of voxels = 45
Object 3: number of voxels = 45
Object 4: number of voxels = 66
Object 5: number of voxels = 159
Object 6: number of voxels = 45
Object 7: number of voxels = 45
Object 8: number of voxels = 58
Object 9: number of voxels = 276
Object 10: number of voxels = 45
Object 11: number of voxels = 138
Object 12: number of voxels = 45
Object 13: number of voxels = 45
Object 14: number of voxels = 137
Object 15: number of voxels = 45
Woolz can also be used as a set of C libraries, or via a binding to another language. Currently these bindings exist for Java, Python and R. For details of these bindings see the JavaWoolz, PyWoolz and RWoolz repositories.
The encoding of Woolz objects when serialised to files is defined only by the source code in the WlzReadObj()/WlzWriteObj() functions. For historical reasons there is no unique identifier (magic number) for Woolz objects. Woolz objects can also (usualy with some loss of information) be written to other (external) file formats.
The program WlzExtFFConvert can be used to convert between supported file formats. As with all the small Woolz programs, the -h option will show usage, but for WlzExtFFConvert it will also list the file formats which are understood:
prompt% WlzExtFFConvert -h
Usage: WlzExtFFConvert [-h] [-s] [-b<background>]
[-d<min-dimension>] [-D<max-dimension>]
[-f<input format>] [-F<output format>]
[-x<x size>] [-y<y size>] [-z<z size>]
[-o<output file>] [<input file>)]
Converts objects between one file format and another, neither of
which need be the Woolz data file format.
Version: 1.8.2
Options:
-h Help, prints this usage information.
-s Split labeled volumes into domains.This will also split
a tiled or pyramidal tiff into resolution slices and tiles.
-G Apply grey value transforms.
-S Use spatial transforms to create WLZ_TRANS_OBJ objects
(by default just offsets are applied).
-b# Set background to value,
-d# Set size of minimum dimension, i.e. min of width or height
-D# Set size of maximum dimension, i.e. max of width or height
-f# Input file format.
-F# Ouput file format.
-o# Output file name.
-x# X voxel/pixel size.
-y# Y voxel/pixel size.
-z# Z voxel/pixel size.
The known file formats are:
Description Extension
*********** *********
Amira Lattice am
Microsoft Bitmap bmp
Stanford Density den
Netgen neutral mesh format emt
Graphics Interchange Format gif
ANALYZE HDR hdr
Image Cytometry Standard ics
IPLab ipl
JPEG jpg
JSON encoded Woolz jsn
Pascal Frey's medit tetrahedral mesh format mesh
Neuroimaging Informatics Technology Initiative nii
Jonathan Shewchuk's mesh format node
Utah nearly raw raster data (NRRD) format nrrd
Wavefront obj
BioRad Confocal pic
Riken PLY2 ply2
PNM pnm
Drishti dot NC format pvl.nc
Raw raw
SLC slc
GRUMMP SMESH smesh
Stereolithography format stl
Tiff tif
Text txt
Sunvision VFF vff
GRUMMP VMESH vmesh
Visualization Toolkit VTK vtk
Simple example:
WlzExtFFConvert -f wlz -F slc <in.wlz >out.slc
Converts the Woolz object in.wlz to an SLC data file out.slc
More complex example:
WlzExtFFConvert -f den -F pnm -o out.pgm in.den
Converts the Stanford density file in.den to a series of PGM files
each with a name of the form out000001.pgm where the number
encodes the image plane and a control file which specifies the
volume origin, size and voxel dimensions.
By default objects are read from the standard input and written to
the standard output.
File formats which use more than one file can not be read or written
using the standard input or standard output.
The TIFF file format must be read/written from/to a file i.e. not
from/to stdin or stdout
Not all formats can retain position information i.e. they can
only keep the size of the bounding box. In these formats the
size of the bounding box is maintained but the position is set to
(0,0). This implies that conversion back to woolz will, in general,
result in a shifted image, i.e. registration is lost. Most 3D
formats encode this data, of the 2D formats only woolz can retain
all offsets, TIFF can only encode positive offsets.
The library functions WlzEffReadObj() and WlzEffWriteObj() (in libWlzExtFF) can also be used to read and write non-Woolz format files.
Woolz uses Doxygen (http://www.doxygen.org) for documentation, although the best documentation is (as always) the source code itself, the Doxygen documentation is available from https://ma-tech.github.io/Woolz/documentation/html_Core/index.html .
The best way to understand Woolz is through the source code, but what follows attempts to give an overview and may be some help.
typedefstruct_WlzObject
{
WlzObjectTypetype;
intlinkcount;
WlzDomaindomain;
WlzValuesvalues;
WlzPropertyList*plist;
struct_WlzObject*assoc;
} WlzObject;The fields encode the type, link count, spatial domain, values, properties and any associated objects of an object. The type simply encodes what the object is (3D image, 2D polygon, ...). Object use reference counting via the linkcount makes many Woolz operations, such as thresholding, extremely efficient with regard to both space and time. The link (reference) count of an object is incremented when an object is assigned using WlzAssignObject() and decremented when an object is freed using WlzFreeObj(). Typical usage looks like:
WlzObject*obj;
obj=WlzAssignObject(WlzReadObj(stdin, &errNum), NULL);
/* Do something with obj */
(void )WlzFreeObj(obj);The domain of an object is the spatial extent within which the object is defined, some other spatial representation (including meshes), some transformation or bizarrely a histogram.
The values of an object are some values, such as intensities, which are only defined within the objects domain.
There is a core Woolz object type:
typedefstruct_WlzCoreObject
{
WlzObjectTypetype;
intlinkcount;
} WlzCoreObject;which is sufficient to determine the type of and object and to either assign it or free it. So, "there's inheritance too, sort of". The other type of top level Woolz object is an array of objects:
typedefstruct_WlzCompoundArray
{
WlzObjectTypetype;
intlinkcount;
WlzObjectTypeotype;
intn;
WlzObject**o;
WlzPropertyList*plist;
WlzObject*assoc;
} WlzCompoundArray;This is similar to the main Woolz object type but has an array of objects rather than a domain and values (variants of this compound object exist including variants with linked lists).
The domain of an object is (in most cases) the spatial description of and object. For 2D or 3D domain objects, such as the EMAGE anatomy domains or the EMAGE model embryos, the domain encodes the region of space that the anatomical or embryo component occupies. Just as there is a core object there is a core domain which has the same uses as the core object:
typedefstruct<_WlzCoreDomain
{
WlzObjectTypetype;
intlinkcount;
void*freeptr;
} WlzCoreDomain;The additional field (compared to WlzCoreObject) is the free pointer. This is a pointer to a stack of memory blocks that have been allocated for the domain. The Woolz Domain is a union of the possible domains:
typedefunion_WlzDomain
{
struct_WlzCoreDomain*core;
struct_WlzIntervalDomain*i;
struct_WlzPlaneDomain*p;
struct_WlzPolygonDomain*poly;
struct_WlzBoundList*b;
struct_WlzHistogramDomain*hist;
struct_WlzRect*r;
struct_WlzFRect*fr;
struct_WlzAffineTransform*t;
struct_WlzWarpTrans*wt;
struct_WlzContour*ctr;
struct_WlzMeshTransform*mt;
struct_WlzLBTDomain2D*l2;
struct_WlzLBTDomain3D*l3;
struct_WlzCMesh2D*cm2;
struct_WlzCMesh2D5*cm2d5;
struct_WlzCMesh3D*cm3;
struct_WlzPoints*pts;
struct_WlzLUTDomain*lut;
struct_WlzThreeDViewStruct*vs3d;
} WlzDomain;These include domains for 2D and 3D spatial regions (WlzIntervalDomain and WlzPlaneDomain) as well as polygons, boundaries, contours, meshes and transforms (such as affine, basis function and mesh transforms). Transforms are domains since they are spatial mappings. Histograms are domains too (for historical reasons).
The values of an object (again in most cases) represent the actual values that are embedded in the space defined by the domain. An object with a domain but without values is perfectly valid and are frequently used that way, eg to represent some anatomical region. The core values type is:
typedefstruct_WlzCoreValues
{
WlzObjectTypetype;
intlinkcount;
} WlzCoreValues;The Woolz Values is (similarly to the Woolz Domain) a union of the possible Woolz values:
typedefunion_WlzValues
{
struct_WlzCoreValues*core;
struct_WlzRagRValues*v;
struct_WlzRectValues*r;
struct_WlzIntervalValues*i;
struct_WlzConvHullValues*c;
struct_WlzVoxelValues*vox;
struct_WlzObject*obj;
struct_WlzFeatValues*fv;
struct_WlzRectFeatValues*rfv;
struct_WlzIndexedValues*x;
struct_WlzTiledValues*t;
struct_WlzLUTValues*lut;
} WlzValues;The types of values include image values (WlzRagRValues, WlzRectValues, WlzIntervalValues and WlzVoxelValues), convex hulls, indexed values (used with meshes) look up tables and features. The top level Woolz object itself is also a valid values! This allows objects to be defined which include a spatial mapping, such as an affine transform, without computing the transformation of the object.
Property lists are arbitrary lists of properties that are associated with an object such as a meaningful name. They were once used extensively, but have now been reinstated as genuine linked lists. Properties may be used to record the history of an object or object names.
Given that Woolz objects are complex data structures, access functions are required to use their elements. These access methods are for spatial domain objects with values (images) but simple adaptations can make the code applicable to spatial domain objects without values.
Processing 2D Woolz domain objects is most efficiently done by scanning through objects using blocks of contiguous pixels.
WlzGreyPgP,
WlzObject*obj;
WlzIntervalWSpaceiWsp;
WlzGreyWSpacegWsp;
WlzErrorNumerrNum=WLZ_ERR_NONE;
errNum=WlzInitGreyScan(obj, &iWsp, &gWsp);
while((errNum==WLZ_ERR_NONE) &&
((errNum=WlzNextGreyInterval(&iWsp)) ==WLZ_ERR_NONE))
{
gP=gWsp.u_grintptr;
switch(gWsp.pixeltype)
{
caseWLZ_GREY_INT:
for(iPos=iWsp.lftpos; iPos<=iWsp.rgtpos; ++iPos)
{
*(gP.inp+iPos) /= 2;
}
break;
default:
errNum=WLZ_ERR_GREY_TYPE;
break;
}
}
if(errNum==WLZ_ERR_EOO)
{
errNum=WLZ_ERR_NONE;
}There are random access functions for establishing whether some vertex within a spatial domain object (WlzInsideDomain()) and the value at some position in space, either inside or outside the object:
intval;
WlzIVertex3pos;
WlzObject*obj;
WlzGreyValueWSpace*gVWSp;
WlzErrorNumerrNum=WLZ_ERR_NONE;
gVWSp=WlzGreyValueMakeWSp(obj, &errNum);
if(errNum==WLZ_ERR_NONE)
{
WlzGreyValueGet(gVWSp, pos.vtZ, pos.vtYpos.vtX);
switch(gVWSp->gType)
{
caseWLZ_GREY_INT:
val= (*(gVWSp->gVal)).inv;
break;
default:
errNum=WLZ_ERR_GREY_TYPE;
break;
}
}Simple iterators are another way to access the values of spatial domain objects in scan order. Unlike WlzNextGreyInterval() the iterators work for both 2D and 3D. In the example below all (integer) values of the spatial domain object are incremented from 0.
WlzErrorNumSetInvGreyValues(WlzObject*obj)
{
inti=0;
WlzIterateWSpace*itWSp=NULL;
WlzErrorNumerrNum=WLZ_ERR_NONE;
itWSp=WlzIterateInit(obj, WLZ_RASTERDIR_IPILIC, 1, &errNum);
if(errNum==WLZ_ERR_NONE)
{
while((errNum=WlzIterate(itWSp)) ==WLZ_ERR_NONE)
{
*(itWSp->gP.inp) =i++;
}
if(errNum==WLZ_ERR_EOO)
{
errNum=WLZ_ERR_NONE;
}
}
WlzIterateWSpFree(itWSp);
return(errNum);
} Woolz geometric models provide a unified representation of both 2D and 3D geometric models composed of simplices. They are capable of simple planar straight line graphs in 2D and both manifold and non-manifold surfaces (with non intersecting elements) in 3D.
Woolz geometric models are used to represent contours such as iso-value contours extracted from domain objects.
An affine transform is a transformation which preserves lines and the parallelism of lines, but not necessarily lengths or the angles between (non parallel) lines. Affine transforms in Woolz are stored as homogeneous 3x3 and 4x4 matrices (actually 4x4 arrays are used for both but 2D transforms access them as if they are 3x3).
Basis function transforms allow displacements at discrete points to be interpolated throughout an object's domain. These are the sum of radially symmetric component transforms, with each component transform centred on one of the points.
Data structures and methods exist within Woolz for both convex and conforming meshes composed of simplices in 2, 2.5 (2D topology but 3D geometry, ie surfaces in 3D space) and 3D. Transforms may be built using the meshes by associating displacements with the mesh nodes, in the case of conforming meshes this is done using indexed values.