Images - Kernel sizes and ImageAxes

General Usage
1

Hi All,

This is probably a naive question. Given an image stack with z-spacing that is not the same as x and y, what is the correct way to handle the size for example of the σshape for the laplacian of gaussian? I have always compensated for that in the size of the filter… i.e., σshape = [1.0, 1.0, 0.5] - but I’m wondering if I should instead be leveraging ImageAxes to set the units, and then the σshape is simply [1,1,1] ? I guess I’m confused on how the setting the units for the axes affects the sizes for the filters.

Thanks in advance for any guidance!

2

Image filtering doesn’t take axis sizing into account:

julia> img = AxisArray(zeros(3, 5),
                       Axis{:x}(1mm:2mm:5mm),
                       Axis{:y}(1mm:1mm:5mm)); img[2, 3] = 1;

julia> img
2-dimensional AxisArray{Float64,2,...} with axes:
    :x, (1:2:5) mm
    :y, (1:5) mm
And data, a 3×5 Array{Float64,2}:
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  1.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0

julia> imfilter(img, Kernel.LoG((0.25, 0.25)))
2-dimensional AxisArray{Float64,2,...} with axes:
    :x, (1:2:5) mm
    :y, (1:5) mm
And data, a 3×5 Array{Float64,2}:
 0.0  0.000137553    0.191352  0.000137553  0.0
 0.0  0.191352     -81.4873    0.191352     0.0
 0.0  0.000137553    0.191352  0.000137553  0.0

It’d be nice if it did, though! I just opened an issue for this: https://github.com/JuliaImages/ImageFiltering.jl/issues/176

1 Like
3

It’s well set up to make this easy, but we might need a @require or two to make it automatic. You should store your array with “rich” axis information using an AxisArray. JuliaImages’ pixelspacing trait then makes things easy.

Here’s a quick demo:

julia> using ImageAxes, AxisArrays, Unitful

julia> using Unitful: μm

julia> img = AxisArray(rand(10, 10, 4), (:y, :x, :z), (0.3μm, 0.3μm, 2μm))
3-dimensional AxisArray{Float64,3,...} with axes:
    :y, (0.0:0.3:2.7) μm
    :x, (0.0:0.3:2.7) μm
    :z, (0:2:6) μm
And data, a 10×10×4 Array{Float64,3}:
[:, :, 1] = # truncated for brevity

julia> pixelspacing(img)
(0.3 μm, 0.3 μm, 2 μm)

julia> σ = 10μm ./ pixelspacing(img)
(33.333333333333336, 33.333333333333336, 5.0)
2 Likes
4

Thank you both for the answers. Super helpful!

1 Like