Update to newer version of Images

Closes #39

Project.toml

@@ -1,6 +1,6 @@
 name = "ImageTracking"
 uuid = "7632190a-71f5-11e9-04ef-99e081817785"
-version = "0.2.0"
+version = "0.3.0"
 
 [deps]
 AxisAlgorithms = "13072b0f-2c55-5437-9ae7-d433b7a33950"
@@ -14,8 +14,8 @@ Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 [compat]
 AxisAlgorithms = "0.3, 1"
 CoordinateTransformations = "0.5, 0.6"
-Images = "0.18, 0.19, 0.20, 0.21, 0.22, 0.23"
-Interpolations = "0.10, 0.11, 0.12, 0.13"
+Images = "0.25, 0.26"
+Interpolations = "0.10, 0.11, 0.12, 0.13, 0.14"
 StaticArrays = "0.10, 0.11, 0.12, 1"
 julia = "1"
 

src/haar.jl

@@ -170,7 +170,7 @@ M. Oren, C. Papageorgiou, P. Sinha, E. Osuna and T. Poggio, "Pedestrian detectio
 P. Viola and M. Jones, "Rapid object detection using a boosted cascade of simple features," Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001, 2001, pp. I-511-I-518 vol.1.
 
 """
-function haar_features(img::AbstractArray{T, 2}, top_left::Array{I, 1}, bottom_right::Array{I, 1}, feat::Symbol, coordinates = nothing) where {T <: Union{Real, Color}, I <: Int}
+function haar_features(img::IntegralArray{T, 2}, top_left::Array{I, 1}, bottom_right::Array{I, 1}, feat::Symbol, coordinates = nothing) where {T <: Union{Real, Color}, I <: Int}
     check_coordinates(top_left, bottom_right)
     rectangular_feature_coordinates = check_feature_type(feat)
 
@@ -198,7 +198,7 @@ function haar_features(img::AbstractArray{T, 2}, top_left::Array{I, 1}, bottom_r
             ybot = top_left[1] + a[j, 3] - 1
             xtop = top_left[2] + a[j, 2] - 1
             xbot = top_left[2] + a[j, 4] - 1
-            rectangular_features[i,j] = boxdiff(img, ytop, xtop, ybot, xbot)
+            rectangular_features[i,j] = img[ytop .. ybot, xtop .. xbot]
         end
     end
     rectangular_feature_values = (sum(rectangular_features[:, 2:2:end], dims = 2) - sum(rectangular_features[:, 1:2:end], dims = 2))[:]

src/lucas_kanade.jl

@@ -1,12 +1,12 @@
-struct LKPyramid
-    layers::Vector{Matrix{Gray{Float64}}}
+struct LKPyramid{T, A<:AbstractMatrix{T}}
+    layers::Vector{A}
 
-    Iy::Union{Nothing, Vector{Matrix{Gray{Float64}}}}
-    Ix::Union{Nothing, Vector{Matrix{Gray{Float64}}}}
+    Iy::Union{Nothing, Vector{A}}
+    Ix::Union{Nothing, Vector{A}}
 
-    Iyy::Union{Nothing, Vector{Matrix{Gray{Float64}}}}
-    Ixx::Union{Nothing, Vector{Matrix{Gray{Float64}}}}
-    Iyx::Union{Nothing, Vector{Matrix{Gray{Float64}}}}
+    Iyy::Union{Nothing, Vector{IntegralArray{T, 2, A}}}
+    Ixx::Union{Nothing, Vector{IntegralArray{T, 2, A}}}
+    Iyx::Union{Nothing, Vector{IntegralArray{T, 2, A}}}
 end
 
 function LKPyramid(image, levels; downsample = 2, σ = 1.0, compute_gradients::Bool = true)
@@ -15,14 +15,15 @@ function LKPyramid(image, levels; downsample = 2, σ = 1.0, compute_gradients::B
         return LKPyramid(pyramid, nothing, nothing, nothing, nothing, nothing)
 
     total_levels = levels + 1
-    Iy = Vector{Matrix{Gray{Float64}}}(undef, total_levels)
-    Ix = Vector{Matrix{Gray{Float64}}}(undef, total_levels)
+    M = typeof(first(pyramid))
+    Iy = Vector{M}(undef, total_levels)
+    Ix = Vector{M}(undef, total_levels)
 
-    Iyy = Vector{Matrix{Gray{Float64}}}(undef, total_levels)
-    Ixx = Vector{Matrix{Gray{Float64}}}(undef, total_levels)
-    Iyx = Vector{Matrix{Gray{Float64}}}(undef, total_levels)
+    Iyy = Vector{IntegralArray{eltype(M), 2, M}}(undef, total_levels)
+    Ixx = Vector{IntegralArray{eltype(M), 2, M}}(undef, total_levels)
+    Iyx = Vector{IntegralArray{eltype(M), 2, M}}(undef, total_levels)
 
-    filling = Fill(zero(eltype(pyramid[1])))
+    filling = Fill(zero(eltype(M)))
     for (i, layer) in enumerate(pyramid)
         Iy[i], Ix[i] = imgradients(layer, KernelFactors.scharr, filling)
         Iyy[i], Ixx[i], Iyx[i] = compute_partial_derivatives(Iy[i], Ix[i])
@@ -158,27 +159,28 @@ function compute_partial_derivatives(Iy, Ix; σ = 4)
 
     squared .= Iy .* Iy
     imfilter!(filtered, squared, kernel_factors)
-    Iyy_integral_table = integral_image(filtered)
+    Iyy_integral_table = IntegralArray(filtered)
 
     squared .= Ix .* Ix
     imfilter!(filtered, squared, kernel_factors)
-    Ixx_integral_table = integral_image(filtered)
+    Ixx_integral_table = IntegralArray(filtered)
 
     squared .= Iy .* Ix
     imfilter!(filtered, squared, kernel_factors)
-    Iyx_integral_table = integral_image(filtered)
+    Iyx_integral_table = IntegralArray(filtered)
 
     Iyy_integral_table, Ixx_integral_table, Iyx_integral_table
 end
 
 function _compute_spatial_gradient(
     grid, Iyy_integral, Iyx_integral, Ixx_integral,
 )
-    sum_Iyy = boxdiff(Iyy_integral, grid[1], grid[2])
-    sum_Ixx = boxdiff(Ixx_integral, grid[1], grid[2])
-    sum_Iyx = boxdiff(Iyx_integral, grid[1], grid[2])
+    sum_Iyy = Iyy_integral[makeiv(grid[1]), makeiv(grid[2])]
+    sum_Ixx = Ixx_integral[makeiv(grid[1]), makeiv(grid[2])]
+    sum_Iyx = Iyx_integral[makeiv(grid[1]), makeiv(grid[2])]
     SMatrix{2, 2, Float64}(sum_Iyy, sum_Iyx, sum_Iyx, sum_Ixx)
 end
+makeiv(range::AbstractUnitRange) = first(range) .. last(range)
 
 function compute_spatial_gradient(pyramid::LKPyramid, grid, level)
     G = _compute_spatial_gradient(

test/haar.jl

@@ -52,7 +52,7 @@ using Images, StaticArrays, LinearAlgebra
     # Simple Image haar_features
 
     img = Matrix(Diagonal([1,1,1,1,1]))
-    int_img = integral_image(img)
+    int_img = IntegralArray(img)
 
     test_features = haar_features(int_img, [2,2], [4,4], :x2)
     correct_features = [-1,0,0,-1,1,1,-1,0,0,1]