The relevant declaration in include/lsst/afw/image/Pixel.h:73
template <typename _ImagePixelT, typename _MaskPixelT, typename _VariancePixelT = double>
class SinglePixel : public detail::MaskedImagePixel_tag {
public:
template <typename, typename, typename>
friend class Pixel;
template <typename T>
friend class PixelTypeTraits;
typedef _ImagePixelT ImagePixelT;
typedef _MaskPixelT MaskPixelT;
typedef _VariancePixelT VariancePixelT;
...
convolution multiplies and adds SinglePixel instances include/lsst/afw/math/ConvolveImage.h:274
inline typename OutImageT::SinglePixel convolveAtAPoint
...
typename OutImageT::SinglePixel outValue = 0;
...
outValue += *inImageLocator * kVal;
include/lsst/afw/image/Pixel.h:604
template <typename ExprT1, typename ExprT2>
BinaryExpr<ExprT1, ExprT2, std::multiplies<typename exprTraits<ExprT1>::ImagePixelT>,
bitwise_or<typename exprTraits<ExprT1>::MaskPixelT>,
variance_multiplies<typename exprTraits<ExprT1>::VariancePixelT> >
operator*(ExprT1 e1, ExprT2 e2) {
return BinaryExpr<ExprT1, ExprT2, std::multiplies<typename exprTraits<ExprT1>::ImagePixelT>,
bitwise_or<typename exprTraits<ExprT1>::MaskPixelT>,
variance_multiplies<typename exprTraits<ExprT1>::VariancePixelT> >(e1, e2);
}
For all (+,-,*,/) pixel operations, the mask operation is bitwise_or. Actually we do not have any operations with bitwise_and.
Proposal:
I'd propose a variant of the convolution operation that performs bitwise_and at the image and kernel pixel multiplication step. So that the kernel as a MaskedImage could carry a region "AND-mask" which would be bitwise_or-ed around a result pixel by the addition step. In this way, regions depending on the kernel value could be defined that determine the mask value of the output pixel.
We (Meredith, Eric, John) discussed this on 2019-01-29 and thought it probably makes sense to assign this to Gabor, although we don't think it should be his immediate priority.