Morphology
Description
Morphological
operations process an image based on the shape of a
structuring element — a small binary matrix, represented in GIL by a
kernel_2d, that is compared against the neighborhood of each pixel. They
are most commonly applied to binary (black and white) images, for example to
remove noise or extract shape features, but GIL’s implementation works on
grayscale channel values too.
morphology.hpp defines two base operations, dilate and erode, and a
number of operations built out of them by chaining them or subtracting their
results.
Morphological operations expect a binary image. Use
threshold_binary
to convert a grayscale image first.
|
Base operations
dilate replaces each pixel with the maximum value found within the
structuring element’s neighborhood; it grows bright regions and shrinks dark
ones. erode replaces each pixel with the minimum value found in the
neighborhood, having the opposite effect. Both accept an iterations
parameter to apply the operation repeatedly.
// dilate(src_view, dst_view, structuring_element, iterations)
dilate(view(img), view(img_out), ker_mat, 1);
erode(view(img), view(img_out), ker_mat, 1);Compound operations
| Operation | Description |
|---|---|
|
Erosion followed by dilation. Removes small bright spots (noise) while leaving the overall shape and size of larger bright regions largely intact. |
|
Dilation followed by erosion, the opposite of opening. Closes small dark holes inside bright regions. |
|
Difference between the dilation and the erosion of the image. Highlights the outline of objects in the image. |
|
Difference between the image and its opening. Extracts small bright elements and details that opening would otherwise remove. |
|
Difference between the closing of the image and the image itself. Extracts small dark elements and details that closing would otherwise remove. |
// opening(src_view, dst_view, structuring_element)
opening(view(img), view(img_out), ker_mat);
closing(view(img), view(img_out), ker_mat);
// morphological_gradient/top_hat/black_hat(src_view, dst_view, structuring_element)
morphological_gradient(view(img), view(img_out), ker_mat);
top_hat(view(img), view(img_out), ker_mat);
black_hat(view(img), view(img_out), ker_mat);Structuring element
The structuring element is a kernel_2d of 0`s and `1`s: only the pixels
overlapped by a `1 participate in the min/max comparison. A common choice is
a square of `1`s, which makes every pixel in the neighborhood participate:
std::vector<float> ker_vec(9, 1.0f); // 3x3 structuring element, all ones
gil::detail::kernel_2d<float> ker_mat(ker_vec.begin(), ker_vec.size(), 1, 1);Demo
See the full example at example/morphology.cpp, which applies each operation from the command line to a PNG image.