https://svn.lrde.epita.fr/svn/oln/branches/cleanup-2008/milena/sandbox Index: ChangeLog from Ugo Jardonnet <ugo.jardonnet@lrde.epita.fr> INIM: Clean up classification. * classif/iccvg04.cc: Clean it up. iccvg04.cc | 82 ++++++++----------------------------------------------------- 1 file changed, 11 insertions(+), 71 deletions(-) Index: classif/iccvg04.cc --- classif/iccvg04.cc (revision 2688) +++ classif/iccvg04.cc (working copy) @@ -14,7 +14,6 @@ #include <mln/util/array.hh> #include <mln/labeling/compute.hh> -// FIXME: !? #include <mln/geom/all.hh> #include <mln/io/ppm/load.hh> @@ -99,86 +98,45 @@ void classify_image(const I& ima, const J& histo, const K& ws, int nbasins, int f) { - unsigned count[nbasins + 1]; memset(count, 0, (nbasins + 1) * sizeof(unsigned)); - algebra::vec<3, unsigned> sum[nbasins + 1]; + algebra::vec<3, unsigned> sum[nbasins + 1]; for (int i = 1; i < nbasins + 1; ++i) sum[i] = literal::zero; + // Compute representatives of every class mln_piter(I) p(ima.domain()); for_all(p) { point3d p3(ima(p).red() / f, ima(p).green() / f, ima(p).blue() / f); int w = ws(p3); - // Check if we are not on a border of the WS + //check if we are not on a border of the WS if (w != 0) { count[w] += histo(p3); sum[w] += histo(p3) * convert::to< algebra::vec<3, value::int_u8> >(p3); } -#if DEBUG - std::cerr << "p3 : " << p3 << " == " << convert::to<algebra::vec<3, value::int_u8> >(p3) << std::endl; -#endif // DEBUG } - for (int i = 1; i < nbasins + 1; ++i) - { - sum[i] *= f; - sum[i] /= count[i]; - //std::cout << "count[" << i << "] = " << count[i] << std::endl; -#if DEBUG - std::cout << "sum[" << i << "] = " << sum[i] << std::endl; -#endif // DEBUG - } - + sum[i] = (sum[i] * f) / count[i]; + // Make an output image where colors are replaced by their representatives. mln_piter(I) pi(ima.domain()); I out(ima.domain()); - I ws_out(ima.domain()); - for_all(pi) { - value::rgb8 coul = ima(pi); + //retrieve color class + value::rgb8 coul = ima(pi); int w = ws(point3d(coul.red() / f, coul.green() / f, coul.blue() / f)); - if (w != 0) // If w == 0 it means that the current point is part of a border of the watershed - { -#if DEBUG - std::cerr << "out(" << pi << ") = sum[" << w << "]; //" << sum[w] << std::endl; -#endif + //if w == 0, out(pi) = 0 ie is part of a border of the watershed out(pi) = convert::to<value::rgb8>(sum[w]); - //out(pi) = value::rgb8(sum[w][0], sum[w][1], sum[w][2]); - ws_out(pi) = value::rgb8(0, 0, 0); } - else - { -#if DEBUG - std::cerr << "Border : " << pi << std::endl; -#endif - ws_out(pi) = value::rgb8(255, 255, 255); - - // FIXME - // ima(pi) is a border in the histogram and therefore lacks of a mean color. - // Choosing a good value for out(pi) is not so easy. One idea could be to randomly - // choose a color from one of the surrounding classes, but it's not the best idea - // because of large « border » zones. - out(pi) = value::rgb8(255, 255, 255); - } - - //std::cerr << "(" << coul << ") -> (" << out(pi) << "); sum[" << w << "] = " << sum[w] << " -> " << convert::to<value::rgb8>(sum[w]) << std::endl; - } - - // 1- La valeur contenue dans sum n'est pas correctement écrite dans out.ppm. À vérifier ! - // 2- Le cast de sum[w] en rgb8 se passe très bizarrement. À vérifier ! - // 3- Rajoute le std::cout de count change le comportement du programme. À vérifier ! - // 4- Le WS semble faire des pâtés (beaucoup de points de classe 0) io::ppm::save(out, "out.ppm"); - io::ppm::save(ws_out, "out-ws.ppm"); } bool usage(int argc, char ** argv) @@ -196,7 +154,6 @@ { if (not usage(argc, argv)) return 1; - const int div_factor = atoi(argv[2]); const int lambda = atoi(argv[3]); @@ -205,36 +162,19 @@ //make histo image3d<unsigned> histo = fill_histo(ima,div_factor); - { - unsigned m, M; - estim::min_max(histo, m, M); - std::cout << "histo : " << m << ' ' << M << std::endl; - } //revert histo image3d<unsigned> rhisto = arith::revert(histo); - { - unsigned m, M; - estim::min_max(rhisto, m, M); - std::cout << "rhisto : " << m << ' ' << M << std::endl; - } - //compute closing_area of histo + //compute closing_volume of histo image3d<unsigned> histo_closure(histo.domain()); - morpho::closing_area(rhisto, c6(), lambda, histo_closure); - { - unsigned m, M; - estim::min_max(histo_closure, m, M); - std::cout << "histo_closure : " << m << ' ' << M << std::endl; - } + morpho::closing_volume(rhisto, c6(), lambda, histo_closure); //watershed over histo_closure unsigned nbasins = 0; image3d<unsigned> ws = morpho::meyer_wst(histo_closure, c6(), nbasins); - std::cout << "nbassins : " << nbasins << std::endl; - // Classify image ! - //classify_image(const I& ima, const J& histo, const K& ws, int nbasins, int f) + //classify image classify_image(ima, histo, ws, nbasins, div_factor); }