<machine Learning in Action > 42-point K-Means algorithm C # to achieve image segmentation

Using system;using system.drawing;using system.windows.forms;using system.threading;using  System.collections.generic;namespace k_means{public partial class Form1:form {const int picturesize = 640        * 480;                             const int k = 3;        The number of centroid color[] Arraycolor = new Color[picturesize];   Data set int[,] resultarray = new int[picturesize, 2];    Store the result of the cluster allocation for each point and the distance to the centroid int[,] Centarray = new int[k, 4];        Storage centroid int[,] Tmpcentarray = new int[k, 4];            Public Form1 () {InitializeComponent ();        Picturebox1.imagelocation = "Test.jpg"; } private void Button1_Click (object sender, EventArgs e) {Bitmap map = new Bitmap (picturebox1.im            Age); for (int j = 0; J < map. Height; J + +) {for (int i = 0; i < map. Width; i++) {arraycolor[j * map. Width + i] = map.                GetPixel (i, j);           } } New Thread (new ThreadStart (Kmeans)).        Start (); private void Button1_resize (object sender, EventArgs e) {this.            Width = 800; This.        Height = 600; } private void Button2_Click (object sender, EventArgs e) {Bitmap map = new Bitmap (picturebox1.im            Age);            Bitmap newmap = map;            Color color = new color ();            Color Newcolor = new color ();            Byte R, G, B, gray; for (int j = 0; J < map. Height; J + +) {for (int i = 0; i < map. Width; i++) {color = map.                    GetPixel (i, j); R = Color.                    R g = Color.                    G b = Color.                    B if (r + G + b! = 0) {Gray = (Byte) ((R * 19595 + G * 38469 + b * 7472) >>                        16);                        Newcolor = Color.FromArgb (Gray, Gray, gray); Newmap. SeTpixel (i, J, Newcolor);        }}} pictureBox1.Image = Newmap; } private int Getdistance (int[,] A, Color b, int i) {return (int) math.sqrt (Math.pow (a[i, 0]-B.        R, 2) + Math.pow (a[i, 1]-B.G, 2) + Math.pow (A[i, 2]-b.b, 2) + Math.pow (A[i, 3]-B.A, 2));            } private void Kmeans () {DateTime dt = DateTime.Now; for (int i = 0; i < K; i++)//initial k random centroid, color value does not look at data set, direct 0-255 {tmpcentarray[i, 0] = centarray[ I, 0] = new Random (Bitconverter.toint32 (GUID.NEWGUID). Tobytearray (), 0)).                Next (0, 256); Tmpcentarray[i, 1] = centarray[i, 1] = new Random (Bitconverter.toint32 (Guid.NewGuid (). Tobytearray (), 0)).                Next (0, 256); Tmpcentarray[i, 2] = centarray[i, 2] = new Random (Bitconverter.toint32 (Guid.NewGuid (). Tobytearray (), 0)).                Next (0, 256); Tmpcentarray[i, 3] = centarray[i, 3] = new Random (Bitconverter.toint32 (GUid. NewGuid (). Tobytearray (), 0)).            Next (0, 256);                 } while (true) {for (int i = 0; i < arraycolor.length; i++)//Traverse data Set each point and cluster allocation                    {int mindistance = getdistance (Centarray, Arraycolor[i], 0);                    int count = 0; for (int j = 1; j < K; J + +) {int tmpdistance = getdistance (Centarray, ARRAYC                        Olor[i], J);                            if (Mindistance > tmpdistance) {mindistance = tmpdistance;                        Count = j;                    }} resultarray[i, 0] = count;                Resultarray[i, 1] = mindistance;                    } for (int j = 0; J < 3; J + +)//calculate new centroid {int r = 0;                    int g = 0;                    int b = 0;    int a = 0;                int m = 0;                        for (int i = 0; i < arraycolor.length; i++) {if (resultarray[i, 0] = = j) {r + = Arraycolor[i].                            R G + = Arraycolor[i].                            G B + = Arraycolor[i].                            B A + = Arraycolor[i].                            A                        m++;                         }} if (M! = 0) {centarray[j, 0] = r/m;                        Centarray[j, 1] = g/m;                        Centarray[j, 2] = b/m;                    Centarray[j, 3] = a/m; } this. Invoke (new MethodInvoker () = {This.label1.Text = centarray[j, 0] + "\ r \ n" + centarray[j, 1] + "\ r \ n" + centarray[j, 2] + "\ r \ n" + centarray[j, 3] + "\ r \ n" + (DATETIME.NOW-DT). TotalSeconds;                })); } bool Isbreak= true;                    for (int i = 0; i < K; i++)//detect if the centroid is still changed {for (int j = 0; J < 4, j + +)                            {if (tmpcentarray[i, j]! = Centarray[i, j]) {                            Isbreak = false;                        Break                    }}} if (Isbreak)//center of mass is not changed, you can exit {                Break                    } else {for (int i = 0; i < K; i++)//Save the calculated centroid of the previous round {for (int j = 0; J < 4; J + +) {Tmpcentarra                        Y[i, j] = Centarray[i, j]; }}}}} private void Button3_Click (object sender, EventArgs E {Bitmap map = new Bitmap (picturebox1.width, picturebox1.height);            for (int j = 0; J < map. Height; J + +) {for (int i = 0; i < map. Width; i++) {if (RESULTARRAY[J * map. Width + I, 0] = = 0) {map.                    SetPixel (i, J, color.red); } else if (RESULTARRAY[J * map. Width + I, 0] = = 1) {map.                    SetPixel (i, J, Color.green); } else if (RESULTARRAY[J * map. Width + I, 0] = = 2) {map.                    SetPixel (i, J, Color.yellow);        }}} pictureBox1.Image = map; }    }}

Code casually write, no optimization, the two points K-means some details have not been clear, had to first use the general K-means do, the effect is to see below it.

<machine Learning in Action > 42-point K-Means algorithm C # to achieve image segmentation