Adds Gamma Correction Function

[mronian]

No description provided.

[coveralls]

Coverage decreased (-1.03%) to 74.695% when pulling 6f3100e on mronian:gamma-correction into f50b32b on timholy:master.

[timholy]

Any time you pass type information via a string, you're going to have type-instability. I let this slide on the histogram equalization PR, but we should get away from that pattern ASAP. It's much better to pass the type as a type.

[bjarthur]

what are we going to do with imstretch, which performs "symmetric gamma-correction" ?

[coveralls]

Coverage decreased (-0.4%) to 74.416% when pulling 8ec10eb on mronian:gamma-correction into 919f480 on timholy:master.

[mronian]

@timholy I have made the function pixel-wise and the code has reduced by a lot (just a line for color images, the chain of functions is a bit long though). Like histeq, the dtype causes the return type to be marked red with @code_warntype. Also, map() seem to be marked red ( they return a Union, eg Union{Array{Float64,2},Array{Union{},2}}). Is this an error?

[coveralls]

Coverage decreased (-0.4%) to 74.416% when pulling 044ec95 on mronian:gamma-correction into 919f480 on timholy:master.

[timholy]

This line is one source of type instability. It can be replicated with

The type instability can be replicated with
```jl
function prep_table{D}(::Type{D}, gamma_inv)
    raw_type = FixedPointNumbers.rawtype(D)
    [D((i / typemax(raw_type)) ^ gamma_inv) for i in zero(raw_type):typemax(raw_type)]
end

But put a D in front of the comprehension, and suddenly inference is happier. Probably JuliaLang/julia#15276.

[timholy]

If you're interested in pursuing this further, it would be good to address @bjarthur's question too.

[mronian]

@bjarthur We could have a function adjust_gamma which takes in a method for symmetric gamma correction as an argument. The imstretch function has matrix dot style operations and needs to be changed anyways to a pixel wise one. Would it be better to do it as a part of adjust_gamma or a separate function is better?

[mronian]

Made a lot of changes, simplified adjust_gamma on the lines of #516 and tests for the same. Tests fail because isapprox for ARGB, AGray is not merged/tagged(JuliaGraphics/ColorVectorSpace.jl#47).

[mronian]

There is still some type instability issue with map-

julia> function foo(i)
           i*i
       end
foo (generic function with 1 method)

julia> img = reshape(1:1:100, 10, 10)
10×10 Base.ReshapedArray{Int64,2,StepRange{Int64,Int64},Tuple{}}:
  1  11  21  31  41  51  61  71  81   91
  2  12  22  32  42  52  62  72  82   92
  3  13  23  33  43  53  63  73  83   93
  4  14  24  34  44  54  64  74  84   94
  5  15  25  35  45  55  65  75  85   95
  6  16  26  36  46  56  66  76  86   96
  7  17  27  37  47  57  67  77  87   97
  8  18  28  38  48  58  68  78  88   98
  9  19  29  39  49  59  69  79  89   99
 10  20  30  40  50  60  70  80  90  100

julia> @code_warntype map(foo, img)
Variables:
  #self#::Base.#map
  f::#foo
  A::Base.ReshapedArray{Int64,2,StepRange{Int64,Int64},Tuple{}}

Body:
  begin  # abstractarray.jl, line 1478:
      return $(Expr(:invoke, _collect(::Base.ReshapedArray{Int64,2,StepRange{Int64,Int64},Tuple{}}, ::Base.Generator{Base.ReshapedArray{Int64,2,StepRange{Int64,Int64},Tuple{}},#foo}, ::Base.EltypeUnknown, ::Base.HasShape), :(Base._collect), :(A), :($(Expr(:new, Base.Generator{Base.ReshapedArray{Int64,2,StepRange{Int64,Int64},Tuple{}},#foo}, :(f), :(A)))), :($(Expr(:new, :(Base.EltypeUnknown)))), :($(Expr(:new, :(Base.HasShape))))))
  end::Union{Array{Int64,2},Array{Union{},2}}

The end line is marked red.

[timholy]

That's been one of the most long-standing problems in julia: the possibility that the input would be an empty array made type-stable map impossible, because map needed at least one element in order to compute the type of the output.

Having been frustrated by this issue for at least 2 years now (e.g., JuliaLang/julia#8027), I'm very pleased to report that...if you update your julia, you'll see this has (very recently) been fixed! There were some deep-guts improvements that made it finally possible to use type-inference to reliably infer the output type of foo even if there were no actual elements to play with.

[timholy]

YIQ

[timholy]

Aside from this tiny edit, LGTM. Seems to need a rebase before merging, however.

[mronian]

Done :) Tests fail due to isapprox.