Different fonts #9

Activity

1. ericphanson commented on Aug 1, 2020

   

   ericphanson

   on Aug 1, 2020

   Owner

   More actions

   Cool, I'm glad you're interested! That looks like a great way. I briefly tried to do it via FreeTypeAbstraction.jl but I got lost in how that package works. The following seems to work!

   using Luxor, Colors, StaticArrays
   
   function string_to_matrix(str)
       m = @imagematrix begin
           background("black")
           sethue("white")
           fontface("Georgia")
           fontsize(24) # warning: big font size -> big matrix -> slow distance
           text(str, halign=:center, valign=:middle)
       end
       sa = SparseArrays.sparse(convert.(Colors.Gray, m) .> convert(Colors.Gray, ARGB32(0.0, 0.0, 0.0, 1.0)))
       return sa
   end
   
   using UnbalancedOptimalTransport, SparseArrays
   
   # `UnbalancedOptimalTransport` uses `DiscreteMeasure` to store the things it is comparing.
   # These have two relevant pieces: the density, which in this case is the opacity or weight we want to put on each pixel,
   # and the `set`, which here will just be the coordinates of the matrix.
   function UnbalancedOptimalTransport.DiscreteMeasure(sa::SparseMatrixCSC)
       coords = SVector{2,Float64}[]
       density = Float64[]
       I, J, V = findnz(sa)
       for n = eachindex(I, J, V)
           push!(coords, SVector{2, Float64}(I[n], J[n]))
           push!(density, V[n])
       end
       return DiscreteMeasure(density, coords)
   end
   
   function visual_distance2(str1, str2; D = UnbalancedOptimalTransport.KL(1.0), ϵ = 0.01)
       sa1 = string_to_matrix(str1)
       sa2 = string_to_matrix(str2)
       return sinkhorn_divergence!(D, DiscreteMeasure(sa1), DiscreteMeasure(sa2), ϵ)
   end

   One difficulty is if the image gets too big, visual_distance gets very slow; with font size 24 it seems OK. I actually don't use VisualStringDistances here at all, just the underlying algorithm in UnbalancedOptimalTransport.jl. Let me know if you have any luck with this or run into any troubles :)
2. cormullion commented on Aug 2, 2020

   

   cormullion

   on Aug 2, 2020

   Author

   More actions

   Thanks, this works great - I can get all sorts of measurements from comparing strings and fonts...

   julia> # test - should be zero
   compare("O", "O", font1="JuliaMono-Regular", font2="JuliaMono-Regular")
   O in JuliaMono-Regular  vs O in JuliaMono-Regular:
   0.0010819308549554307
   
   julia> # compare different glyphs in the same font
   compare("O", "0", font1="JuliaMono-Regular", font2="JuliaMono-Regular")
   O in JuliaMono-Regular  vs 0 in JuliaMono-Regular:
   26.77476852981517
   
   julia> compare("O", "0", font1="SFMono-Regular", font2="SFMono-Regular")
   O in SFMono-Regular  vs 0 in SFMono-Regular:
   39.771611214031
   
   julia> # compare different glyphs in the same font
   compare("&", "\$", font1="JuliaMono-Regular", font2="JuliaMono-Regular")
   & in JuliaMono-Regular  vs $ in JuliaMono-Regular:
   71.03373098044722
   
   julia> compare("&", "\$", font1="SFMono-Regular", font2="SFMono-Regular")
   & in SFMono-Regular  vs $ in SFMono-Regular:
   78.41157442985032
   
   julia> # comparing same glyphs in different fonts
   compare("0", "0", font1="JuliaMono-Regular", font2="SFMono-Regular")
   0 in JuliaMono-Regular  vs 0 in SFMono-Regular:
   23.482812287382792
   
   julia> compare("O", "O", font1="JuliaMono-Regular", font2="SFMono-Regular")
   O in JuliaMono-Regular  vs O in SFMono-Regular:
   15.58672859276412
   

   Literally hours of fun... :)

   By the way, you can add dimensions to the matrix to improve performance, the default is 250×250:

   m = @imagematrix begin
               background("black")
               sethue("white")
               fontface(font)
               fontsize(24) # warning: big font size -> big matrix -> slow distance
               text(str, halign=:center, valign=:middle)
           end 30 30
   

   Thanks!

   PS: Sinkhorn Divergence would be a good name for my new band... :)

   🎉React with 🎉1Reacted by Eric Hanson