Content-aware image cropping with ChunkyPNG

README.md

Simplest possible solution to image cropping problem:

• grab the most "exciting" part of the image
  • "exciting" => least average according to unweighed color
• checks every single 100 X 100 splotch, so it's very inefficient. Not checking every single subimage would be the most obvious improvement, but frankly, doing image processing in native ruby is probably not the best idea anyhow.
• special bonus! also finds the least interesting 100X100 subimage.
• usage: ruby chunky.rb <inputfn.png>
• most "exciting" crop is saved as <inputfn.png>.max.png
• most "boring" crop is saved as <inputfn.png>.min.png

This works:

• extremely well for the duck
• well for the cat
• so-so for the dog

Actually in the case of the dog, the bit identified as "most boring" is a pretty good crop. Reason being that the dog picture has little variation: half big black dog, half green. All said, though, the dog crop is acceptable and by far not the worst possible choice.

Possible improvements:

• tons of performance optimizations
• calculating the averages differently, especially perception weighting the individual color channels or hsv components.
• doing this properly would require more sophisticated algorithms, I doubt it would be feasible with ChunkyPNG

cat.png.max.png

cat.png.min.png

chunky.rb

require 'chunky_png'


include ChunkyPNG


def usage
  STDERR.puts "usage: [bla] filename.png"
  exit 1
end

usage if 1 != ARGV.length 


# determine "average" color of a splotch.
def avg img
  sum = img.pixels.inject(0) {|sum, pixel|
    sum += pixel
  }  
  sum/img.pixels.length
end

# iterate over every possible 100 X 100 
# subimage
def each img
  # replace `upto` with `step` here to 
  # speed things up
  0.upto(img.width-100) {|x|
    0.upto(img.height-100) {|y|
      yield img.crop(x,y,100,100)
    }
  }
end



img     = Image.from_file(ARGV[0])
img_avg = avg(img)



max              = 0
min              = 0xffffffffffff
img_max, img_min = nil, nil

each(img) { |i|
  d = (avg(i)-img_avg).abs
  max, img_max = d, i if d > max
  min, img_min = d, i if d < min
}



img_max.to_image.save(ARGV[0]+".max.png")
img_min.to_image.save(ARGV[0]+".min.png")

dog.png.max.png

dog.png.min.png

duck.png.max.png

duck.png.min.png

Nice results, simple code. I like this solution.