Gray arithmetic

[bjarthur]

when mixing Gray with Number the type of the result depends on the order of the inputs:

julia> using ColorVectorSpace, Colors

julia> 2.0 / Gray(2.0)
1.0

julia> Gray(2.0) / 2.0
Gray{Float64}(1.0)

julia> Gray(2.0) / Gray(2.0)
1.0

my feeling is that the results should always be Gray, but wanted to ask for opinions before submitting a PR to fix it, as a whole a lot has to be changed, for all of + and - always returns Number when the inputs are of mixed types:

julia> Gray(2.0)+Gray(2.0)
Gray{Float64}(4.0)

julia> Gray(2.0)+2.0
4.0

julia> 2.0+Gray(2.0)
4.0

[timholy]

For the record, here's the logic behind the current choices:

• In adding/subtracting quantities, at one point I felt it would be best to convert to plain numbers unless everything were gray. Not sure about this at all.
• In Gray/Gray, the grays "cancel," much as units like meter cancel when you perform division. Stated differently, if you're computing a ratio of image intensities, isn't this a number rather than a grayscale value?
• The weirdest one is 2.0/Gray(2.0). We don't have a type with units 1/Gray, so I made it just default to a real value.

I don't even begin to pretend I know how to make the right call here. This is murky territory, so I invite debate.

[bjarthur]

so long as there is a logic, i think i can live with it. thanks.

[timholy]

I do wonder if we should change addition and subtraction, though.

[bjarthur]

addition and subtraction treat Array{Gray} and Gray in a way which doesn't seem consistent:

julia> data = convert(Array{Gray{Float64}},rand(3))
3-element Array{ColorTypes.Gray{Float64},1}:
 Gray{Float64}(0.750356)
 Gray{Float64}(0.671796)
 Gray{Float64}(0.595428)

julia> data - 0.1
3-element Array{ColorTypes.Gray{Float64},1}:
 Gray{Float64}(0.650356)
 Gray{Float64}(0.571796)
 Gray{Float64}(0.495428)

julia> data[1] - 0.1
0.6503562677627358

[timholy]

Base julia uses an interesting rule: when performing an operation on an array and a scalar, the type of the array wins.

For example:

julia> typeof(UInt8(5)+3)
Int64

julia> typeof([UInt8(5)]+3)
Array{UInt8,1}

This was chosen to prevent promotion from blowing up the size required to store the result.

[johnnychen94]

from users' experience, what can we expect about the output type of some complicated function?

"Colorants in, Colorants out" satisfies me most

[timholy]

There's real merit in that. The current system may be too complicated to intuit.

A defense of the current system, though, might be

meanval = mean(img)     # this would be Gray
ratio = img ./ meanval  # this would be Float64

To me this makes a bit of sense, if I'm computing a ratio I expect a plain number.

Anyway, I'm open to changing this.

[timholy]

Fixed by #131