public override Output Execute(System.Drawing.Bitmap input, bool debug = false, bool showSteps = false)
{
Output o = new Output();
RgbByteArrayBitmap rgb = RgbByteArrayBitmap.FromBitmap((input));
//progressBarSetMax(100); // we are so fast, we dont need no progress bar(we are also not showing debug or steps)
//int step = rgb.Length / 100;
for (int i = 0; i < rgb.Length; i++)
{
rgb[i] = (byte)(255 - rgb[i]);
//if ((i % step) == 0)
// progressBarStep();
}
o.Add(rgb.ToBitmap(), "negative");
return(o);
}
/// <summary>
/// Creates a new 3-channel instance copied its values from a provided byte bitmap
/// </summary>
public DoubleArrayBitmap(RgbByteArrayBitmap rgbBytes)
{
Width = rgbBytes.Width;
Height = rgbBytes.Height;
Channels = new double[3][];
for (int j = 0; j < 3; j++)
{
Channels[j] = new double[Width * Height];
}
for (int i = 0; i < Width * Height; i++)
{
Channels[0][i] = rgbBytes[i * 3];
Channels[1][i] = rgbBytes[i * 3 + 1];
Channels[2][i] = rgbBytes[i * 3 + 2];
}
}
/// <summary>
/// Applies the algorithm to a bitmap
/// </summary>
public override Output Execute(Bitmap input, bool debug = false, bool showSteps = false)
{
// create an output object to later add output (and debug) images to
Output o = new Output();
// convert the input to an RgbByteArrayBitmap for faster processing
RgbByteArrayBitmap rgb = RgbByteArrayBitmap.FromBitmap((input));
// set preview image, anouncing that processing has begone
setPreviewBitmap((Bitmap)input.Clone(), "processing..", Color.Red);
// calculate the number of radii later tested and their relative spacing(f)
int min_r = 10;
int radii = Math.Min(Math.Min(rgb.Width, rgb.Height) / 4, 40);
double f = Math.Max(1, (Math.Min(rgb.Width, rgb.Height) / 6.0 - min_r) / radii);
// set progressbar maximum
progressBarSetMax(radii * 2);
// create a dilation kernel
Kernel max = new KernelMax();
max.AddCircleElements(2.5, 1);
// create a median kernel
Kernel median = new KernelMedian();
median.IgnoreBorders = false;
median.AddCircleElements(1.5, 1);
// create a simple averaging kernel
Kernel smooth = new Kernel();
smooth.AddCircleElements(1.5, 1);
smooth.IgnoreBorders = false;
// create a x-derivative kernel
Kernel dx = new Kernel();
dx.AddElement(-1, 0, 1);
dx.AddElement(1, 0, -1);
// create a y-derivative kernel
Kernel dy = new Kernel();
dy.AddElement(0, -1, 1);
dy.AddElement(0, 1, -1);
// set up some variables
DoubleArrayBitmap result;
DoubleArrayBitmap resultDx;
DoubleArrayBitmap resultDy;
// convert byte array bitmap to hsv
result = rgb.ToHSV();
// extract brightness channel
DoubleArrayBitmap brightness = result.ExtractChannel(2);
// if preview is enabled, show the brightness channel
if (showSteps)
{
setPreviewBitmap(brightness.ToBytes().ToBitmap(), "grey values", Color.Red);
}
// taking median and smoothing brightness to reduce noise
brightness = median.ApplyToBitmap(brightness);
brightness = smooth.ApplyToBitmap(brightness);
// show progress
if (showSteps)
{
setPreviewBitmap(brightness.ToBytes().ToBitmap(), "reduced noise", Color.Red);
}
// add brightness image to debug output
if (debug)
{
o.Add(brightness.ToBytes().ToBitmap(), "debug: brightness");
}
// calculate derivatives
resultDx = dx.ApplyToBitmap(brightness);
resultDy = dy.ApplyToBitmap(brightness);
// square derivatives
resultDx.SquareMe();
resultDy.SquareMe();
// calculate gradient
DoubleArrayBitmap gradient = resultDx + resultDy;
gradient.SqrtMe();
// show progress
if (showSteps)
{
setPreviewBitmap(gradient.ToBytes().ToBitmap(), "gradient");
}
// dilate gradient slightly, improves detection
gradient = max.ApplyToBitmap(gradient);
// show progress
if (showSteps)
{
setPreviewBitmap(gradient.ToBytes().ToBitmap(), "dilated gradient");
}
// add gradient to debug output
if (debug)
{
o.Add(gradient.ToBytes().ToBitmap(), "debug: gradient");
}
// reduce size of gradient image, making detection faster by a factor of 4
gradient = gradient.ShrinkByTwo();
// initializing array to hold detected features for different radii
DoubleArrayBitmap[] steps = new DoubleArrayBitmap[radii];
// detect features for the different radii
for (int r = 0; r < radii; r++)
{
// update progressbar
progressBarStep();
// create 'special' 24-point clockface kernel(see makeClockFaceKernel-method)
Kernel clockface = makeClockFaceKernel(min_r + (int)(r * f));
// apply the kernel to detect features
steps[r] = clockface.ApplyToBitmap(gradient);
// square the resulting image to make both negative and positive features stand out
steps[r].SquareMe();
// show progress
if (showSteps)
{
setPreviewBitmap(steps[r].ExpandByTwo().ToBytes().ToBitmap(), "pattern detection " + r + "/" + radii);
}
}
// find maximum over all features in xyr-space
double maximum = 0;
for (int r = 0; r < radii; r++)
{
for (int i = 0; i < gradient.Width * gradient.Height; i++)
{
if (steps[r][0, i] > maximum)
{
maximum = steps[r][0, i];
}
}
}
// initialize another array of bitmaps to hold the final smoothed features
DoubleArrayBitmap[] steps_final = new DoubleArrayBitmap[radii];
// create a new bitmap to sum up all steps_final for debug output
if (debug)
{
result = new DoubleArrayBitmap(gradient.Width, gradient.Height);
}
// initialize 3d-array to hold thresholded object information extracted from detected features
bool[,,] xyrSpace = new bool[gradient.Width, gradient.Height, radii];
// for all radii...
for (int r = 0; r < radii; r++)
{
int a = Math.Max(r - 2, 0);
int b = Math.Min(r + 2, radii);
steps_final[r] = new DoubleArrayBitmap(gradient.Width, gradient.Height);
// sum up detected features over up to 5 r-layers ( smoothes features )
for (int x = 0; x < gradient.Width * gradient.Height; x++)
{
for (int i = a; i < b; i++)
{
steps_final[r][0, x] += steps[i][0, x];
}
}
// theshold r-layer if needed for preview/debug-output
if (showSteps || debug)
{
steps_final[r].ThresholdMe(maximum);
}
// show smoothed and thresholded preview of r-layer
if (showSteps)
{
setPreviewBitmap(steps_final[r].ExpandByTwo().ToBytes().ToBitmap(), "threshold " + r + "/" + radii);
}
// threshold r-layer and store in xyrSpace as boolean
for (int x = 0; x < gradient.Width; x++)
{
for (int y = 0; y < gradient.Height; y++)
{
xyrSpace[x, y, r] = steps_final[r][0, x, y] > maximum;
}
}
// add thresholded r-layer to debug output
if (debug)
{
result += steps_final[r];
o.Add(steps_final[r].ExpandByTwo().ToBytes().ToBitmap(), "debug: radius " + (min_r + (int)(r * f)).ToString());
}
// update progress bar
progressBarStep();
}
// debug output sum of thresholded r-layers
if (debug)
{
o.Add(result.ExpandByTwo().ToBytes().ToBitmap(), "debug: sum(all radii)"); // debug
}
// boolean-xyrSpace is filled now!
// initialzed a list of detected clock faces
ClockFaceList clList = new ClockFaceList();
//clList.AllowOverlap = true; // enabling this will detect many smaller features that are not really clock faces
// search through xyrSpace from large to small radius
for (int r = radii - 1; r >= 0; r--)
{
for (int x = 0; x < gradient.Width; x++)
{
for (int y = 0; y < gradient.Height; y++)
{
if (xyrSpace[x, y, r])
{
// if object is found, extract and add to list
int[] center = extractObject(ref xyrSpace, x, y, r);
// perform some funny computations to offset previous resolution change and take into acount radius factor f
clList.Add(center[0] * 2, center[1] * 2, (int)((min_r + center[2] * f) * 2 * 1.1));
}
}
}
}
// create final output bitmaps
// "output" shows all detected clock faces in different gray values on black
// "mask" shows the original image masked with the detected clock faces, leaving everything else black
Bitmap output = new Bitmap(input.Width, input.Height);
Graphics g = Graphics.FromImage(output);
Bitmap mask = new Bitmap(input);
Graphics g_m = Graphics.FromImage(mask);
g.Clear(Color.Black);
g_m.Clear(Color.Black);
int clId = clList.ClockFaces.Count;
foreach (ClockFace cl in clList.ClockFaces)
{
int c = clId * 255 / clList.ClockFaces.Count;
Brush b = new SolidBrush(Color.FromArgb(c, c, c));
Point p = new Point(cl.X - cl.R, cl.Y - cl.R);
Size s = new Size(cl.R * 2, cl.R * 2);
g.FillEllipse(b, new Rectangle(p, s));
b = new SolidBrush(Color.White);
g_m.FillEllipse(b, new Rectangle(p, s));
clId--;
}
// add final outputs to output list
o.Add(output, "detected clock faces");
// also perform bitwise AND operation with input bitmap here to get masked image
o.Add((RgbByteArrayBitmap.FromBitmap(mask) & rgb).ToBitmap(), "masked input");
return(o);
}
