I have the following piece of code:
public List<Tuple<double, double, double>> GetNormalizedPixels(Bitmap image)
{
System.Drawing.Imaging.BitmapData data = image.LockBits(
new Rectangle(0, 0, image.Width, image.Height),
System.Drawing.Imaging.ImageLockMode.ReadOnly,
image.PixelFormat);
int pixelSize = Image.GetPixelFormatSize(image.PixelFormat) / 8;
var result = new List<Tuple<double, double, double>>();
unsafe
{
for (int y = 0; y < data.Height; ++y)
{
byte* row = (byte*)data.Scan0 + (y * data.Stride);
for (int x = 0; x < data.Width; ++x)
{
Color c = Color.FromArgb(
row[x * pixelSize + 3],
row[x * pixelSize + 2],
row[x * pixelSize + 1],
row[x * pixelSize]);
// (*)
result.Add(Tuple.Create(
1.0 * c.R / 255,
1.0 * c.G / 255,
1.0 * c.B / 255);
}
}
}
image.UnlockBits(data);
return result;
}
The key fragment (*) is this:
result.Add(Tuple.Create(
1.0 * c.R / 255,
1.0 * c.G / 255,
1.0 * c.B / 255);
which adds a pixel with its components scaled to range [0, 1]
to be further used in classification tasks with different classifiers. Some of them require the attributes to be normalized like this, others don't care - hence this function.
However, what should I do when I'd like to classify pixels in a different colour space than RGB
, like L*a*b*
? While values of all coordinates in RGB
colour space fall into range [0,256)
in L*a*b*
colour space a*
and b*
are said to be unbounded.
So when changing the fragment (*) to:
Lab lab = c.ToLab();
result.Add(Tuple.Create(
1.0 * lab.L / 100,
1.0 * lab.A / ?,
1.0 * lab.B / ?);
(ToLab
is an extension method, implemented using appropriate algorithms from here)
what should I put for the question marks?
In practice the number of all possible RGB
colours is finite, so the L*a*b*
space is bounded. It is easy to find the ranges of coordinates with the following simple program:
Color c;
double maxL = double.MinValue;
double maxA = double.MinValue;
double maxB = double.MinValue;
double minL = double.MaxValue;
double minA = double.MaxValue;
double minB = double.MaxValue;
for (int r = 0; r < 256; ++r)
for (int g = 0; g < 256; ++g)
for (int b = 0; b < 256; ++b)
{
c = Color.FromArgb(r, g, b);
Lab lab = c.ToLab();
maxL = Math.Max(maxL, lab.L);
maxA = Math.Max(maxA, lab.A);
maxB = Math.Max(maxB, lab.B);
minL = Math.Min(minL, lab.L);
minA = Math.Min(minA, lab.A);
minB = Math.Min(minB, lab.B);
}
Console.WriteLine("maxL = " + maxL + ", maxA = " + maxA + ", maxB = " + maxB);
Console.WriteLine("minL = " + minL + ", minA = " + minA + ", minB = " + minB);
or a similar one using any other language.
So, CIELAB
space coordinate ranges are as follows:
L in [0, 100]
A in [-86.185, 98.254]
B in [-107.863, 94.482]
and the answer is:
Lab lab = c.ToLab();
result.Add(Tuple.Create(
1.0 * lab.L / 100,
1.0 * (lab.A + 86.185) / 184.439,
1.0 * (lab.B + 107.863) / 202.345);