Method: SPCore::Statistics.correlation

Defined in:
lib/spcore/analysis/statistics.rb

.correlation(image, feature, zero_padding = 0) ⇒ Object

Determines the normalized cross-correlation of a feature with an image.

Normalization is from -1 to 1, where 1 is high correlation, -1 is high correlation (of inverse), and 0 is no correlation.

For autocorrelation, just cross-correlate a signal with itself.

Parameters:

  • image (Array) —

    The values which are actually recieved/measured.

  • feature (Array) —

    The values to be searched for in the image. Size must not be greater than size of image.

  • zero_padding (Fixnum) (defaults to: 0) —

    Number of zeros to surround the image with.

Raises:

  • (ArgumentError) 


35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
 
# File 'lib/spcore/analysis/statistics.rb', line 35

def self.correlation image, feature, zero_padding = 0
  raise ArgumentError, "feature size is > image size" if feature.size > image.size
  
  unless zero_padding == 0
    image = Array.new(zero_padding, 0) + image + Array.new(zero_padding, 0)
  end
  
  feature_mean = feature.inject(0){ |s, x| s + x } / feature.size.to_f
  feature_diff = feature.map {|x| x - feature_mean }
  sx = feature_diff.inject(0){ |s, x| s + x**2 }
  
  data = []
  for i in 0...(1 + image.size - feature.size)
    region = image[i...(i + feature.size)]
    region_mean = region.inject(0){|s,x| s + x } / feature.size.to_f
    region_diff = region.map {|x| x - region_mean }
    sy = region_diff.inject(0){ |s, x| s + x**2 }
    
    if sx == 0 || sy == 0
      if sx == 0 && sy == 0
        data.push 1.0
      else
        data.push 0.0
      end
      
      next
    end
      
    denom = Math.sqrt(sx*sy)
    
    sum = 0
    feature.size.times do |j|
      sum += (region_diff[j] * feature_diff[j])
    end
    
    r = sum / denom
    data.push(r)
  end
  
  return data
end