public static int EucliceanDistance(NewColor c1, NewColor c2)
{
int redDifference = c1.R - c2.R;
int greenDifference = c1.G - c2.G;
int blueDifference = c1.B - c2.B;
return(redDifference * redDifference + greenDifference * greenDifference + blueDifference * blueDifference);
}
public Cluster(NewColor centroid, int index)
{
this.centroid = centroid;
this.id = index;
}
public byte[] KMeans(byte[] buffer, byte[] result, int k)
{
Random rand = new Random();
var colors = new HashSet <NewColor>();
//uniqe colors in an image
for (int i = 0; i < buffer.Length; i = i + 4)
{
NewColor color = new NewColor(buffer[i + 2], buffer[i + 1], buffer[i]);
colors.Add(color);
}
List <NewColor> allColors = colors.ToList();
// limit k to existing number of colors
if (k > colors.Count)
{
k = colors.Count;
}
NewColor[] centroids = new NewColor[k];
NewColor[] avgCentroids = new NewColor[k];
// random initial centroids
for (int i = 0; i < k; i++)
{
centroids[i] = allColors[rand.Next(allColors.Count)];
}
bool changed = true;
int iteration = 0;
//till centroids are changing and maximum no of iteratioins is not achieved
while (changed && iteration < 100)
{
iteration++;
List <Cluster> clusters = new List <Cluster>();
//assign centroid to a cluster
for (int i = 0; i < k; i++)
{
clusters.Add(new Cluster(centroids[i], i));
clusters[i].sumR = centroids[i].R;
clusters[i].sumG = centroids[i].G;
clusters[i].sumB = centroids[i].B;
clusters[i].noOfcolors = 1;
}
// for all available colors assign clusters and calculate the resulting sums of RGB values
for (int col = 0; col < allColors.Count; col++)
{
double minDistance = double.MaxValue;
int clusterIndex = -1;
NewColor color = allColors[col];
for (int c = 0; c < centroids.Length; c++)
{
double distance = NewColor.EucliceanDistance(color, centroids[c]);
if (distance < minDistance)
{
minDistance = distance;
clusterIndex = c;
}
}
for (int cl = 0; cl < clusters.Count; cl++)
{
if (clusters[cl].id == clusterIndex)
{
clusters[clusterIndex].distances.Add(minDistance);
clusters[clusterIndex].sumR += color.R;
clusters[clusterIndex].sumG += color.G;
clusters[clusterIndex].sumB += color.B;
clusters[clusterIndex].noOfcolors += 1;
}
}
}
//for all centroids calculate the avg color in their cluster: if it equals centroid exit the loop,
//otherwise assign the calculated avg and continue loop
for (int j = 0; j < centroids.Length; j++)
{
int numOfColors = clusters[j].noOfcolors;
avgCentroids[j] = new NewColor((clusters[j].sumR / numOfColors), (clusters[j].sumG / numOfColors), (clusters[j].sumB / numOfColors));
if (centroids[j].R == avgCentroids[j].R && centroids[j].G == avgCentroids[j].G && centroids[j].B == avgCentroids[j].B)
{
changed = false;
}
else
{
centroids[j].R = avgCentroids[j].R;
centroids[j].G = avgCentroids[j].G;
centroids[j].B = avgCentroids[j].B;
}
}
}
//for all pixels
for (int i = 0; i < buffer.Length; i = i + 4)
{
NewColor color = new NewColor(buffer[i + 2], buffer[i + 1], buffer[i]);
double minDistance = double.MaxValue;
int clusterIndex = -1;
for (int m = 0; m < k; m++)
{
double distance = NewColor.EucliceanDistance(color, centroids[m]);
if (distance < minDistance)
{
minDistance = distance;
clusterIndex = m;
}
}
result[i + 3] = 255;
result[i + 2] = (byte)centroids[clusterIndex].R;
result[i + 1] = (byte)centroids[clusterIndex].G;
result[i] = (byte)centroids[clusterIndex].B;
}
return(result);
}
