MaskedImage Iterators

(Return to Images)

This code is in maskedImage2.cc. Note that Image iterators work almost identically, the main difference being how pixel values are set.

If you need access to nearby pixels, see the MaskedImage locator tutorial.

#include "lsst/geom.h"
#include "lsst/afw/image/MaskedImage.h"

// Declare the desired MaskedImage type.
// Note: only specific types are supported; for the list of available types
// see the explicit instantiation code at the end of lsst/afw/image/src/MaskedImage.cc
namespace afwImage = lsst::afw::image;
typedef afwImage::MaskedImage<int> ImageT;

int main() {
    // Declare a MaskedImage; its pixels are not yet initialized.
    ImageT img(lsst::geom::Extent2I(10, 6));

    // Initialize all pixels to a given value.
    img = ImageT::Pixel(100, 0x1, 10);

    // Here is a common and efficient way to set all pixels of the image.
    // Note that the end condition is only computed once, for efficiency.
    for (int y = 0; y != img.getHeight(); ++y) {
        for (ImageT::x_iterator ptr = img.row_begin(y), end = img.row_end(y); ptr != end; ++ptr) {
            *ptr = ImageT::Pixel(100, 0x1, 10);

            // Or, if you prefer, you may set image, mask and variance separately with no loss of speed
            ptr.image() = 100;
            ptr.mask() = 0x1;
            ptr.variance() = 10;
        }
    }

    // It is probably slower to compute the end condition each time, as is done here.
    for (int y = 0; y != img.getHeight(); ++y) {
        for (ImageT::x_iterator ptr = img.row_begin(y); ptr != img.row_end(y); ++ptr) {
            *ptr = ImageT::Pixel(100, 0x1, 10);
        }
    }

    // STL-compliant iterators are available.
    // However, they are not very efficient because the image data may not be contiguous
    // so these iterators must test for end-of-row on every increment.
    // (By the way, we do guarantee that an image's row data is contiguous).
    // iterator
    for (ImageT::iterator ptr = img.begin(), end = img.end(); ptr != end; ++ptr) {
        *ptr = ImageT::Pixel(100, 0x1, 10);
    }
    // reverse_iterator
    for (ImageT::reverse_iterator ptr = img.rbegin(), end = img.rend(); ptr != end; ++ptr) {
        *ptr = ImageT::Pixel(100, 0x1, 10);
    }
    // A different way of choosing begin() for use with (inefficient) iterator
    for (ImageT::iterator ptr = img.at(0, 0), end = img.end(); ptr != end; ++ptr) {
        *ptr = ImageT::Pixel(100, 0x1, 10);
    }

    // There is one efficient STL-compliant iterator: "fast_iterator", but it only works for contiguous images
    // (such as newly allocated images). If you attempt to use this on a subimage you will get an exception.
    for (ImageT::fast_iterator ptr = img.begin(true), end = img.end(true); ptr != end; ++ptr) {
        *ptr = ImageT::Pixel(100, 0x1, 10);
    }

    // It is possible to traverse the image by columns instead of by rows,
    // but because the data is row-contiguous, this has awful consequences upon cache performance.
    for (int x = 0; x != img.getWidth(); ++x) {
        for (ImageT::y_iterator ptr = img.col_begin(x), end = img.col_end(x); ptr != end; ++ptr) {
            *ptr = ImageT::Pixel(100, 0x1, 10);
        }
    }

    // If you must traverse the image by columns then consider doing it in batches to improve
    // cache performance, as shown here:
    int x = 0;
    for (; x != img.getWidth() % 4; ++x) {
        for (ImageT::y_iterator ptr = img.col_begin(x), end = img.col_end(x); ptr != end; ++ptr) {
            *ptr = ImageT::Pixel(100, 0x1, 10);
        }
    }
    for (; x != img.getWidth(); x += 4) {
        for (ImageT::y_iterator ptr0 = img.col_begin(x + 0), end0 = img.col_end(x + 0),
                                ptr1 = img.col_begin(x + 1), ptr2 = img.col_begin(x + 2),
                                ptr3 = img.col_begin(x + 3);
             ptr0 != end0; ++ptr0, ++ptr1, ++ptr2, ++ptr3) {
            *ptr0 = *ptr1 = *ptr2 = *ptr3 = ImageT::Pixel(100, 0x1, 10);
        }
    }

    // Save the image to disk
    img.writeFits("foo");

    return 0;
}