Colorfulness

Node.js implementation of colorfulness using node-opencv binder for OpenCV.

Related research studies

• From Rubner 2000, EMD is a good way to compare 2 images
• From Datta 2006 has used a "perfectly colored" image BGR distribution to compare image with EMD This is what they called the colorfulness measure.

Prerequisites

You will need to make node-opencv work on your local machine, so havind, opencv, node, npm.

Histograms calculation

Which image is the most colorfull ? This library will give you the answer in node.js !

Pre-requisites

• opencv

Installation

npm install colorfulness

Example

var colorfulness = require('colorfulness');

colorfulness("example/image.png", function(err, res){  
  // res is a number of colorfullness between 0 (not colorfull) and 1 (colorfull)
});

// or with open cv lib

var cv = require("opencv");
cv.readImage("example/image.png", function(err, im){
  if(err){
    //handle error
  }

  colorfulness({
    image : im
  }, function(err, res){  
    // res is a number of colorfullness between 0 (not colorfull) and 1 (colorfull)
  });
})

Test

npm test

Results

Images

| File | Image | Colorfulness | |---|---|---| | mona.png | | 60% | | car1.jpg | | 69% | | stuff.png | | 72% | | neutral.png | | 100% | | amaro.png | | 90% | | FFFFFF.png | | 56% | | 000000.png | | 49% | | 00FFFF.png | | 57% |

Non-symetric of measure in BGR space

Remark : FFFFFF.png (white image) is more colorful than 000000.png (black image), it is because the cost function is done in the "LUV" color space.

To understand this, let's consider BGR-centers distance cost matrix in LUV_L2 distance space. To simplify my explanation i will use 2x2x2 = 8 BGR cubes (instead of 64 as used in the code);

Cubes centers are

| Cube number |BGR center position | LUV center position | |---|---|---| | cube 0 | [64,64,64] | [69,97,139] | | cube 1 | [64,64,192] | [117,166,160] | | cube 2 | [64,192,64] | [176,57,211] | | cube 3 | [64,192,192] | [193,100,212] | | cube 4 | [192,64,64] | [90,92,46] | | cube 5 | [192,64,192] | [128,136,68] | | cube 6 | [192,192,64] | [182,61,129] | | cube 7 | [192,192,192] | [198,97,139] |

Matrix of distance in LUV space is looks like :

| | cube 0 | cube 1 | cube 2 | cube 3 | cube 4 | cube 5 | cube 6 | cube 7 | SUM | |---|---|---|---|---|---|---|---|---|---| | cube 0 | 0.00 | 0.44 | 0.69 | 0.74 | 0.49 | 0.51 | 0.61 | 0.66 | 4.14 | | cube 1 | 0.44 | 0.00 | 0.69 | 0.58 | 0.71 | 0.50 | 0.65 | 0.55 | 4.12 | | cube 2 | 0.69 | 0.69 | 0.00 | 0.24 | 0.97 | 0.87 | 0.42 | 0.44 | 4.31 | | cube 3 | 0.74 | 0.58 | 0.24 | 0.00 | 1.00 | 0.83 | 0.47 | 0.37 | 4.23 | | cube 4 | 0.49 | 0.71 | 0.97 | 1.00 | 0.00 | 0.32 | 0.65 | 0.73 | 4.87 | | cube 5 | 0.51 | 0.50 | 0.87 | 0.83 | 0.32 | 0.00 | 0.57 | 0.55 | 4.14 | | cube 6 | 0.61 | 0.65 | 0.42 | 0.47 | 0.65 | 0.57 | 0.00 | 0.21 | 3.58 | | cube 7 | 0.66 | 0.55 | 0.44 | 0.37 | 0.73 | 0.55 | 0.21 | 0.00 | 3.51 |

So pure "cube 0"-distribution (corresponding to FFFFFF image) will not be symetric with "cube 7"-distribution (corresponding to 000000 image).

Pure "cube 4"-distribution (corresponding to 00FFFF image), is even more colorful.