How to compare two colors for similarity/differenc

Question

I want to design a program that can help me assess between 5 pre-defined colors which one is more similar to a variable color, and with what percentage. The thing is that I don't know how to do that manually step by step. So it is even more difficult to think of a program.

More details: The colors are from photographs of tubes with gel that as different colors. I have 5 tubes with different colors were each is representative of 1 of 5 levels. I want to take photographs of other samples and on the computer assess to which level that sample belongs by comparing colors, and I want to know that with a percentage of approximation too. I would like a program that does something like this: http://www.colortools.net/color_matcher.html

If you can tell me what steps to take, even if they are things for me to think and do manually. It would be very helpful.

Answer 1

I used this in my android up and it seems satisfactory although RGB space is not recommended:

    public double colourDistance(int red1,int green1, int blue1, int red2, int green2, int blue2)
{
      double rmean = ( red1 + red2 )/2;
    int r = red1 - red2;
    int g = green1 - green2;
    int b = blue1 - blue2;
    double weightR = 2 + rmean/256;
    double weightG = 4.0;
    double weightB = 2 + (255-rmean)/256;
    return Math.sqrt(weightR*r*r + weightG*g*g + weightB*b*b);
}

Then I used the following to get percent of similarity:

double maxColDist = 764.8339663572415;
double d1 = colourDistance(red1,green1,blue1,red2,green2,blue2);
String s1 = (int) Math.round(((maxColDist-d1)/maxColDist)*100) + "% match";

It works well enough.

Answer 2

See Wikipedia's article on Color Difference for the right leads. Basically, you want to compute a distance metric in some multidimensional colorspace. But RGB is not "perceptually uniform", so your Euclidean RGB distance metric suggested by Vadim will not match the human-perceived distance between colors. For a start, Lab* is intended to be a perceptually uniform colorspace, and the deltaE metric is commonly used. But there are more refined colorspaces and more refined deltaE formulas that get closer to matching human perception.

You'll have to learn more about colorspaces and illuminants to do the conversions. But for a quick formula that is better than the Euclidean RGB metric, just do this: assume that your RGB values are in the sRGB colorspace, find the sRGB to Lab* conversion formulas, convert your sRGB colors to Lab*, and compute deltaE between your two Lab* values. It's not computationally expensive, it's just some nonlinear formulas and some multiplies and adds.

Answer 3

The best way is deltaE. DeltaE is a number that shows the difference of the colors. If deltae < 1 then the difference can't recognize by human eyes. I wrote a code in canvas and js for converting rgb to lab and then calculating delta e. On this example the code is recognising pixels which have different color with a base color that I saved as LAB1. and then if it is different makes those pixels red. You can increase or reduce the sensitivity of the color difference with increae or decrease the acceptable range of delta e. In this example I assigned 10 for deltaE in the line that I wrote (deltae <= 10):

<script>   
  var constants = {
    canvasWidth: 700, // In pixels.
    canvasHeight: 600, // In pixels.
    colorMap: new Array() 
          };



  // -----------------------------------------------------------------------------------------------------

  function fillcolormap(imageObj1) {


    function rgbtoxyz(red1,green1,blue1){ // a converter for converting rgb model to xyz model
 var red2 = red1/255;
 var green2 = green1/255;
 var blue2 = blue1/255;
 if(red2>0.04045){
      red2 = (red2+0.055)/1.055;
      red2 = Math.pow(red2,2.4);
 }
 else{
      red2 = red2/12.92;
 }
 if(green2>0.04045){
      green2 = (green2+0.055)/1.055;
      green2 = Math.pow(green2,2.4);    
 }
 else{
      green2 = green2/12.92;
 }
 if(blue2>0.04045){
      blue2 = (blue2+0.055)/1.055;
      blue2 = Math.pow(blue2,2.4);    
 }
 else{
      blue2 = blue2/12.92;
 }
 red2 = (red2*100);
 green2 = (green2*100);
 blue2 = (blue2*100);
 var x = (red2 * 0.4124) + (green2 * 0.3576) + (blue2 * 0.1805);
 var y = (red2 * 0.2126) + (green2 * 0.7152) + (blue2 * 0.0722);
 var z = (red2 * 0.0193) + (green2 * 0.1192) + (blue2 * 0.9505);
 var xyzresult = new Array();
 xyzresult[0] = x;
 xyzresult[1] = y;
 xyzresult[2] = z;
 return(xyzresult);
} //end of rgb_to_xyz function
function xyztolab(xyz){ //a convertor from xyz to lab model
 var x = xyz[0];
 var y = xyz[1];
 var z = xyz[2];
 var x2 = x/95.047;
 var y2 = y/100;
 var z2 = z/108.883;
 if(x2>0.008856){
      x2 = Math.pow(x2,1/3);
 }
 else{
      x2 = (7.787*x2) + (16/116);
 }
 if(y2>0.008856){
      y2 = Math.pow(y2,1/3);
 }
 else{
      y2 = (7.787*y2) + (16/116);
 }
 if(z2>0.008856){
      z2 = Math.pow(z2,1/3);
 }
 else{
      z2 = (7.787*z2) + (16/116);
 }
 var l= 116*y2 - 16;
 var a= 500*(x2-y2);
 var b= 200*(y2-z2);
 var labresult = new Array();
 labresult[0] = l;
 labresult[1] = a;
 labresult[2] = b;
 return(labresult);

}

    var canvas = document.getElementById('myCanvas');
    var context = canvas.getContext('2d');
    var imageX = 0;
    var imageY = 0;

    context.drawImage(imageObj1, imageX, imageY, 240, 140);
    var imageData = context.getImageData(0, 0, 240, 140);
    var data = imageData.data;
    var n = data.length;
   // iterate over all pixels

    var m = 0;
    for (var i = 0; i < n; i += 4) {
      var red = data[i];
      var green = data[i + 1];
      var blue = data[i + 2];
    var xyzcolor = new Array();
    xyzcolor = rgbtoxyz(red,green,blue);
    var lab = new Array();
    lab = xyztolab(xyzcolor);
    constants.colorMap.push(lab); //fill up the colormap array with lab colors.         
      } 

  }

// -----------------------------------------------------------------------------------------------------

    function colorize(pixqty) {

         function deltae94(lab1,lab2){    //calculating Delta E 1994

         var c1 = Math.sqrt((lab1[1]*lab1[1])+(lab1[2]*lab1[2]));
         var c2 =  Math.sqrt((lab2[1]*lab2[1])+(lab2[2]*lab2[2]));
         var dc = c1-c2;
         var dl = lab1[0]-lab2[0];
         var da = lab1[1]-lab2[1];
         var db = lab1[2]-lab2[2];
         var dh = Math.sqrt((da*da)+(db*db)-(dc*dc));
         var first = dl;
         var second = dc/(1+(0.045*c1));
         var third = dh/(1+(0.015*c1));
         var deresult = Math.sqrt((first*first)+(second*second)+(third*third));
         return(deresult);
          } // end of deltae94 function
    var lab11 =  new Array("80","-4","21");
    var lab12 = new Array();
    var k2=0;
    var canvas = document.getElementById('myCanvas');
                                        var context = canvas.getContext('2d');
                                        var imageData = context.getImageData(0, 0, 240, 140);
                                        var data = imageData.data;

    for (var i=0; i<pixqty; i++) {

    lab12 = constants.colorMap[i];

    var deltae = deltae94(lab11,lab12);     
                                        if (deltae <= 10) {

                                        data[i*4] = 255;
                                        data[(i*4)+1] = 0;
                                        data[(i*4)+2] = 0;  
                                        k2++;
                                        } // end of if 
                                } //end of for loop
    context.clearRect(0,0,240,140);
    alert(k2);
    context.putImageData(imageData,0,0);
} 
// -----------------------------------------------------------------------------------------------------

$(window).load(function () {    
  var imageObj = new Image();
  imageObj.onload = function() {
  fillcolormap(imageObj);    
  }
  imageObj.src = './mixcolor.png';
});

// ---------------------------------------------------------------------------------------------------
 var pixno2 = 240*140; 
 </script>

Answer 4

If you have two Color objects c1 and c2, you can just compare each RGB value from c1 with that of c2.

int diffRed   = Math.abs(c1.getRed()   - c2.getRed());
int diffGreen = Math.abs(c1.getGreen() - c2.getGreen());
int diffBlue  = Math.abs(c1.getBlue()  - c2.getBlue());

Those values you can just divide by the amount of difference saturations (255), and you will get the difference between the two.

float pctDiffRed   = (float)diffRed   / 255;
float pctDiffGreen = (float)diffGreen / 255;
float pctDiffBlue   = (float)diffBlue  / 255;

After which you can just find the average color difference in percentage.

(pctDiffRed + pctDiffGreen + pctDiffBlue) / 3 * 100

Which would give you a difference in percentage between c1 and c2.

Answer 5

A simple method that only uses RGB is

cR=R1-R2 
cG=G1-G2 
cB=B1-B2 
uR=R1+R2 
distance=cR*cR*(2+uR/256) + cG*cG*4 + cB*cB*(2+(255-uR)/256)

I've used this one for a while now, and it works well enough for most purposes.

Answer 6

I expect you want to analyze a whole image at the end, don't you? So you could check for the smallest/highest difference to the identity color matrix.

Most math operations for processing graphics use matrices, because the possible algorithms using them are often faster than classical point by point distance and comparism calculations. (e.g. for operations using DirectX, OpenGL, ...)

So I think you should start here:

http://en.wikipedia.org/wiki/Identity_matrix

http://en.wikipedia.org/wiki/Matrix_difference_equation

... and as Beska already commented above:

This may not give the best "visible" difference...

Which means also that your algorithm depends onto your definiton of "similar to" if you are processing images.