Spatial statistics & image analysis

AI-GEOSTATS software & shareware archive

Metasite providing extensive links to software used in geostatistics, geographic information systems, landscape analysis, and related fields. The most commonly used method is kriging, giving optimal prediction and extrapolation of 2- and 3-dimensional unevenly-spaced data, including quantitative estimation of the uncertainty at each point. Among many other products, AI-GEOSTATS includes:

• GEO-EAS Introductory interactive Fortran geostatistics package for spatial analysis in two-dimensions developed by EPA. Provides basic statistics, variograms and kriging, correspondence analysis, cross-validation, and graphing. P
• UNCERT Large package in C designed for hydrogeologists histograms and plots, varograms and cross-variograms, jackknifing uncertainty estimates, kriging, 3-dimensional visualization of scattered data. P
• Generic Mapping Tools (GMT) Collection of 50 UNIX tools for 2- and 3-dimensional datasets with filtering, spline and nearest neighbor estimation, filtering, many projections and contouring. Produces EPS illustrations. P
• Geostatistical Software Library (GSLIB) Fortran code under UNIX for 2- and 3-dimensional geostatistics including many krigings with uncertainties estimated by simulated annealing, cross-validation and jackknifing. PostScript outputs. P
• NEM Clustering and classification of spatial data based on EM Algorithm implementation of a penalized likelihood statistic. Code in C.

MR/1

Sophisticated package for multiscale multiresolution analysis of complex images based on wavelets. Useful for noise modeling (including Poisson), faint feature detection, filtering and thresholding, image deconvolution, blob analysis, image registration, image compression, mathematical morphology operations, FFT, format conversions (including FITS). Written in C++ and IDL. P (commercial, not free, package)

Computer Vision source codes

Impressive metasite with links to many image processing packages including: SatherVision (object oriented environment); SUSAN algorithms (structure-preseving smoothing, edge/corner detection); GSnake (contour modeling, extraction/detection), Object Recognition Toolkit (progressive edge detection); Steerable Pyramid (multiscale, multiorientation wavelet-based image decomposition); MMach (mathematical morphology toolbox within Khoros, with dilation and erosion); KLT (feature tracker for registration of overlapping images); MeasTex (texture classification). Provides links to toolkits within specific environments (KHOROS, X11/Motif, Vista, Tcl/Tk, MatLab, Macintosh, OpenGL), tool libraries for specific image formats (JPEG, TIFF, MPEG), and codes for image synthesis by ray tracing.

TOOLDIAG

Collection of methods in C for statistical pattern recognition through classification of multidimensional continuous features. Includes several classifier paradigms (k-nearest neighbor, radial basis function network, Parzen window, Q* algorithm), feature selection strategies and criteria (estimated minimal error probability, inter-class distance, probabilistic distance), feature extraction (inear discriminant analysis, principal component analysis, Sammon mapping), cross-validation error estimation and more. (P)

VARIOWIN

PC-based package for spatial data analysis and geostatistics including variogram surfaces, measures of spatial continuity, surface fitting, and graphical tools. Also at this site are the DOS packages Geo-EAS and Geostatistical Toolbox.

Geographic Resources Analysis Support System (GRASS)

A large geographic information system (GIS) developed by the US Corps of Engineers. It is freely available, runs on various platforms, and has extensive support information (user and programming manuals, training, FAQs, conferences, etc.). It provides exploratory spatial analysis, semivariogram model fitting, nearest neighbor analysis, interpolation, tests for spatial autocorrelation, and extensive graphics for multivariate spatial (raster, vector or point) data. (P)

Spatial autocorrelations

Highly efficient Matlab and Fortran toolbox for computing simultaneous and conditional spatial autoregressions and mixed regressive spatially autoregressive models, produced by K. Pace of the Dept. of Finance at Lousiana State University.

AUTO

Two-dimensional autocorrelation function of digital images, including Ghostscript graphics. (P)

Gaussian Markov random field simulation

GMRFsim is a group of C-routines for the fast and exact simulations of general Gaussian random fields on a 2-dimensional lattice and general graph, provided by H. Rue of the Norwegian Institute of Technology. It may be useful both to compare with an observed image or as step in building Bayesian spatial models. P

Morphometrics software

Site providing information and software relating to mathematical morphometrics including: multivariate regression of shape and unwarping of images using thin-plate splines; N-dimensional image rotation and superimposition; shape analysis using landmarks; outline detection; analysis of outlines using Fourier analysis, eigenshape analysis and fractal analysis; image enhancements; and more.

Programs for stochastic geometry

Codes by Dietrich Stoyan, author of "Stochastic geometry & its applications" (1995): Voronoi 1.0 for 3-dimensional tessellations, STG 4.1 for the simulation and statistical analysis of several models of stochastic geometry; and programs for stereology.

Statistical pattern recognition, classification, neural networks, perceptrons, machine vision and learning, data mining, and image processing (including medical imaging).

The following are metasites to the vast resources in these fields: Pattern Recognition Group, TU Delft; TOOLDIAG (Brazil); Computer vision software, Carnegie-Mellon; Classification Society of North America; Knowledge Discovery and data mining, Katholieke Univ; and St@tServ Data Mining Software.

SPFAC

Spatial factor analysis of multivariate data (P)

Multiscale analysis

ADS is a sub-package associated with the multivariate analysis system ADE-4 with multiscale aanalysis and graphical display of spatial point patterns based on Ripley's function K(r).

VORONOI

Computes voronoi diagrams and Delaunay triangulations. (P)

Tesselations

Kriging interpolation

Clear explanation, examples and C code for kriging, an interpolation method for estimating a correlated but random continuous spatial surface from irregularly spaced data. It also provides the variogram which gives the uncertainty of the estimated values at every location. Method developed by geologists.

Minimum spanning tree

Calculation, printing and updating the minimum spanning tree of a multivariate dataset. Also here and here. From the journal Applied Statistics.

GCONTR 2-D contour maps

FARB-E-2D 2-D contour maps on rectangular mesh

CON3D 3-D contour maps

Planar Voronoi diagram

ANGLE

One-sample uniformity tests and multi-sample tests on two-dimensional orientation (directional) data.