* milena/mln/world/binary_2d/subsample.hh: Many fixes. Conflicts: milena/ChangeLog milena/mln/world/binary_2d/subsample.hh --- milena/ChangeLog | 12 +++++++- milena/mln/world/binary_2d/subsample.hh | 51 ++++++++++++++++++++++--------- 2 files changed, 47 insertions(+), 16 deletions(-) diff --git a/milena/ChangeLog b/milena/ChangeLog index 9572039..22bee5f 100644 --- a/milena/ChangeLog +++ b/milena/ChangeLog @@ -1,4 +1,3 @@ -<<<<<<< HEAD 2013-09-02 Roland Levillain <roland@lrde.epita.fr> Add parentheses around complex operands of operator `|'. @@ -4770,6 +4769,17 @@ * mln/geom/crop.hh, * mln/geom/crop_without_localization.hh: New. +2013-09-05 Edwin Carlinet <carlinet@lrde.epita.fr> + + Fix some shortcomings in binary subsampling. This function used to + memory leak and makes an unnecessary heap memory + allocation. Moreover, it makes some unnecessary assumptions about + the domain of the image (only worked for 2D images with a domain + starting at (0,0)). + + * milena/mln/world/binary_2d/subsample.hh: Many fixes. + + 2012-06-05 Edwin Carlinet <carlinet@lrde.epita.fr> Fix issues with loading ascii ppm. diff --git a/milena/mln/world/binary_2d/subsample.hh b/milena/mln/world/binary_2d/subsample.hh index 3ea041a..f6baf6c 100644 --- a/milena/mln/world/binary_2d/subsample.hh +++ b/milena/mln/world/binary_2d/subsample.hh @@ -50,7 +50,18 @@ namespace mln namespace binary_2d { - /// Subsample a Boolean image. + /// \brief Subsample a Boolean image. + /// + /// Subsample a binary image by a factor \p n and return a gray + /// level image by averaging. + /// + /// \[ + /// out(p) = \frac{1}{n^2} \sum_{\epsilon=1}^n input(n.(p+pmin) + + /// \langle \epsilon, 1\rangle ) + /// \] + /// + /// Note that even if the input image does not have its origin + /// at (0,0), the output image does so. /// /// \param[in] input A binary image. /// \param[in] n Linear subsampling coefficient. @@ -75,31 +86,41 @@ namespace mln if (n == 0) return data::convert(int_u8(), input); - const bool** ptr = new const bool*[n]; + const unsigned nrows = input.nrows() / n; const unsigned ncols = input.ncols() / n; - image2d<int_u8> output(nrows, ncols); - - const unsigned delta_row = input.delta_offset(down); - for (unsigned row = 0; row < nrows; ++row) + algebra::vec<2, unsigned int> vmin; + algebra::vec<2, unsigned int> vmax; + vmin[0] = 0; + vmin[1] = 0; + vmax[0] = nrows - 1; + vmax[1] = ncols - 1; + point2d pmin(vmin); + point2d pmax(vmax); + + image2d<int_u8> output(box<point2d>(pmin, pmax)); + dpoint2d dp_row(1, 0); + const unsigned delta_row = input.delta_index(dp_row); + + const bool* origin = &input(input.domain().pmin()); + const bool* line_ptr = origin; + for (unsigned row = 0; row < nrows; ++row, line_ptr += delta_row * n) { - ptr[0] = & input(point2d(n * row, 0)); - for (unsigned i = 1; i < n; ++i) - ptr[i] = ptr[i - 1] + delta_row; - for (unsigned col = 0; col < ncols; ++col) + const bool* ptr = line_ptr; + for (unsigned col = 0; col < ncols; ++col, ptr += n) { unsigned count = 0; for (unsigned i = 0; i < n; ++i) - for (unsigned j = 0; j < n; ++j, ++(ptr[i])) - if (*(ptr[i])) + for (unsigned j = 0; j < n; ++j) + { + if (ptr[delta_row * i + j]) ++count; - output(point2d(row, col)) = count * mln_max(int_u8) / n / n; + } + output.at_(row, col) = count * 255 / n / n; } } - delete[] ptr; - return output; } -- 1.7.10.4