Merge pull request #53 from JuliaImages/teh/hoist

More optimizations

Author: zygmuntszpak

src/algorithms/common.jl

@@ -1,25 +1,30 @@
-function transform_density!(img::GenericGrayImage, edges::AbstractArray, cdf::AbstractArray, minval::Union{Real,AbstractGray}, maxval::Union{Real,AbstractGray})
-    first_edge = first(edges)
+function transform_density!(out::GenericGrayImage, img::GenericGrayImage, edges::AbstractRange, newvals::AbstractVector)
+    first_edge, last_edge = first(edges), last(edges)
+    first_newval, last_newval = first(newvals), last(newvals)
     inv_step_size = 1/step(edges)
-    scale = (maxval - minval) / (cdf[end] - first(cdf))
     function transform(val)
         val = gray(val)
-        if val >= edges[end]
-            newval = cdf[end]
+        if val >= last_edge
+            return last_newval
         elseif val < first_edge
-            newval = first(cdf)
+            return first_newval
         else
             index = floor(Int, (val-first_edge)*inv_step_size) + 1
-            newval = cdf[index]
+            @inbounds newval = newvals[index]
+            return newval
         end
-        # Scale the new intensity value to so that it lies in the range [minval, maxval].
-        newval = minval + (newval - first(cdf)) * scale
     end
-    if eltype(img) <: Integer
-        map!(val->ceil(transform(val)), img, img)
-    else
-        map!(transform, img, img)
+    map!(transform, out, img)
+end
+
+function build_lookup(cdf, minval::T, maxval::T) where T
+    first_cdf = first(cdf)
+    # Scale the new intensity value to so that it lies in the range [minval, maxval].
+    scale = (maxval - minval) / (cdf[end] - first_cdf)
+    if T <: Integer
+        return T[ceil(minval + (x - first_cdf) * scale) for x in cdf]
     end
+    return T[minval + (x - first_cdf) * scale for x in cdf]
 end
 
 function construct_pdfs(img::GenericGrayImage, targetimg::AbstractArray, edges::AbstractRange)

src/algorithms/equalization.jl

@@ -104,13 +104,14 @@ imshow(imgeq)
 end
 
 function (f::Equalization)(out::GenericGrayImage, img::GenericGrayImage)
-    edges, histogram = build_histogram(img, f.nbins, minval = f.minval, maxval = f.maxval)
+    minval, maxval = convert(eltype(out), f.minval), convert(eltype(out), f.maxval)
+    edges, histogram = build_histogram(img, f.nbins, minval = minval, maxval = maxval)
     lb = first(axes(histogram,1))
     ub = last(axes(histogram,1))
     N = length(img)
     cdf = cumsum(histogram[lb:ub]/N)
-    out .= img
-    transform_density!(out, edges, cdf, f.minval, f.maxval)
+    newvals = build_lookup(cdf, minval, maxval)
+    transform_density!(out, img, edges, newvals)
 end
 
 function (f::Equalization)(out::AbstractArray{<:Color3}, img::AbstractArray{<:Color3})

src/build_histogram.jl

@@ -188,17 +188,16 @@ function build_histogram(img::GenericGrayImage, edges::AbstractRange)
     Base.has_offset_axes(edges) && throw(ArgumentError("edges must be indexed starting with 1"))
     lb = first(axes(edges,1))-1
     ub = last(axes(edges,1))
-    first_edge = first(edges)
+    first_edge, last_edge = first(edges), last(edges)
     inv_step_size = 1/step(edges)
     counts = fill(0, lb:ub)
-    for val in img
+    @inbounds for val in img
          if isnan(val)
              continue
          else
-            if val >= edges[end]
-                counts[end] += 1
-                continue
-            elseif val < first(edges)
+            if val >= last_edge
+                counts[ub] += 1
+            elseif val < first_edge
                 counts[lb] += 1
             else
                 index = floor(Int, gray((val-first_edge)*inv_step_size)) + 1