module Tutorial::FindDifferences

Overview

Given differences.jpg image:

We can programatically detect most of the differences between these two images:

require "cr-image"
require "cr-image/jpeg"

image = CrImage::RGBAImage.open("differences.jpg")
ret = image.clone

# Perform a blur to remove random noise
img1 = image.gaussian_blur(1)[2...340, ..]
img2 = image.gaussian_blur(1)[361...699, ..]

# Convert each color channel into a IntMap
red_img1 = CrImage::UInt8Map.new(img1.width, img1.red).to_i
gre_img1 = CrImage::UInt8Map.new(img1.width, img1.green).to_i
blu_img1 = CrImage::UInt8Map.new(img1.width, img1.blue).to_i
red_img2 = CrImage::UInt8Map.new(img2.width, img2.red).to_i
gre_img2 = CrImage::UInt8Map.new(img2.width, img2.green).to_i
blu_img2 = CrImage::UInt8Map.new(img2.width, img2.blue).to_i

# Calculate a pixel by pixel difference calculation, convert it to a `Mask`
red_mask = (red_img1 - red_img2).abs.>(50)
gre_mask = (gre_img1 - gre_img2).abs.>(50)
blu_mask = (blu_img1 - blu_img2).abs.>(50)

# Combine the different color channel masks
combo_mask = (red_mask | blu_mask | gre_mask).dilate!

# Expensive way to make a thick box.
3.times do
  combo_mask.dilate!.segments.each do |v|
    # Draw a box in the first image
    ret.draw_square!(v.region.x + 2, v.region.y, v.region.width, v.region.height, CrImage::Color.of("#f00"))
    # ... and the second image
    ret.draw_square!(v.region.x + 361, v.region.y, v.region.width, v.region.height, CrImage::Color.of("#f00"))
  end
end

ret.save("differences_end.jpg")

Results in image:

This script ran on my machine in 66ms in --release mode.

NOTE: There's a difference in the seeds on one of the watermelons that don't survive the blurring process, and so aren't picked up by the differences algorithm above.

Defined in:

tutorials/find_differences.cr