Changes from beta to version 1.0 ================================ Listed below are the major changes which have been made from the beta version to version 1.0. I have listed only those changes which affect the framework and its operation. Other minor changes have occured but do not significantly alter the framework. 1. Added the suffix .ts to all test suite directories in the imgs subdirectory. This allows valid test suites to be easily distinguished from documentation or anciliary files and directories. To update the test suite directories from Beta version to version 1.0, simply append the suffix .ts to existing test suite directory names. Scripts/programs affected: bin/collateResults bin/genAlgHTML bin/runalg bin/runscore bin/runtest bin/runtype imgs/_artBrick/buildBombMortar imgs/_artBrickL/buildLattice 2. Added the suffix .alg to all algorithm result directories in the algs subdirectory. This allows valid algorithm directories to be easily distinguished from documentation or anciliary files and directories. To retain existing results from Beta version, append .alg to all algorithm result directories in MEASTEX_ROOT/algs. Also, for each algorithm directory, append .ts to all test suite subdirectory names. Scripts/programs affected: bin/cleanalg bin/collateResults bin/genAlgHTML bin/genResHTML bin/processAllScores bin/runalg bin/runscore bin/runtest bin/runtype 3. Added three environment variables (MEASTEX_ROOT, MEASTEX_ALGS and MEASTEX_IMGS) to allow more flexibility when running tests. When MEASTEX_ROOT is set, the user no longer needs to be in the top-level directory to run any of the scripts. Also, the MEASTEX_ALGS and MEASTEX_IMGS environment variables allow multiple users to access the framework, saving their own results in a private algorithm directory and accessing individual test suites. The MEASTEX_IMGS directory may just be a directory with links to a subset of the global test suites. See http:www/for_using.html for usage information. Scripts/programs affected: bin/cleanalg bin/collateResults bin/genAlgHTML bin/genResHTML bin/processAllScores bin/processScores bin/runalg bin/runscore bin/runtest bin/runtype 4. Modified performance metric to add normalization option. Scores can now be normalized with respect to the probabilistic prior probabability score or normalized with respect to the binary (winner-take-all) prior probability score. The score can also be left un-normalized. This option must also be handled by runscore, processScores and processAllScores. Scripts/programs affected: bin/scoretest bin/runscore bin/processScores bin/processAllScores src/general/scoretest.c 5. Distribution method for visTex and newTex test suites have been changed to reduce size of distribution. The original images are no longer distributed, just instructions on where and how to obtain the full image files, the .Test and .Valid files, and scripts to extract the subimages. 6. The operation to make the GLCM symmetric did not correctly handle the diagonal elements. This bug has been fixed. NOTE: If you obtained results using the beta version of this algorithm, then they will be invalid. The bug has a reasonably minor effect on the results but an improvement is generally achieved with the correct implementation. Scripts/programs affected: src/glcm/glcmClass.c 7. The set of standard Nth order symmetric masks (N = 1,7) have been added as optional masks for the Gaussian Markov Random Fields algorithm. Scripts/programs affected: src/markov/markovClass.c, markovClass.h 8. Addition feature set options were added to the GLCM algorithm. Three commonly used glcm feature sets were implemented. Scripts/programs affected: src/glcm/glcmClass.c 9. A new test suite, demo.ts has been added to test the installation. 10. A second classifier (K-Nearest Neighbour) has been added to all algorithms to complement the multivariate Gaussian Bayesian classifier Scripts/programs affected: src/glcm/glcmClass.c, glcmClass.h src/gabor/gaborClass.c, gaborClass.h src/markov/markovClass.c, markovClass.h src/lib/Makefile src/lib/KNNClass.c src/lib/MVGClass.c src/lib/MeasTex.h 11. A new algorithm, fractalClass, has been added which implements a classifier using features based on Fractal Dimension of an image. Scripts/programs affected: src/fractal/fractalClass.c, fractalClass.h, Makefile src/Makefile