Conversion to 8 bit

Gwinzy · March 22, 2021, 3:50pm

Hey,
i have a pretty basic question but i dont find a solution for it.
I dont really understand the way ImageJ converts a color jpg to 8 bit.
If i have the values
red = 248, green = 177, blue = 149
the sum is 574 and divded by 3 it is 191.3333
ImageJ makes it 192.

If i have the values
red = 247, green = 176, blue = 148
the sum is 574 and divded by 3 it is 190.3333
ImageJ makes it 190.

This look just random to me, does anyone know how exactly that gets calculated? Is it not just pixel by pixel? All alpha values are 255 by the way.

Best regards
Gwinzy

Research_Associate · March 22, 2021, 4:55pm

I haven’t checked the code, but just off the top of my head I assume it is something like:

Where each step is rounded to 0 decimal places prior to addition.

Bio7 · March 24, 2021, 9:30am

You can find the equations here:

https://imagej.nih.gov/ij/docs/menus/image.html

RGB images are converted to grayscale using the formula gray=(red+green+blue)/3 or gray=0.299red+0.587green+0.114blue if “Weighted RGB to Grayscale Conversion” is checked in Edit>Options>Conversions .

See also:
https://imagejdocu.tudor.lu/gui/image/type

Source:

github.com

imagej/imagej1/blob/32aeb17f323ea102afb5fe1c66332070fd1395fa/ij/process/TypeConverter.java#L95


			bp.setColorModel(ip.getColorModel());
			bp.resetMinAndMax();		//don't take min&max from ip
			return bp;
		}
	}

	/** Converts a ColorProcessor to a ByteProcessor. 
		The pixels are converted to grayscale using the formula
		g=r/3+g/3+b/3. Call ColorProcessor.setRGBWeights() 
		to do weighted conversions. */
	ByteProcessor convertRGBToByte() {
		if (ip.getNChannels()==1 && doScaling) {
			byte[] pixels8 = ip.create8BitImage();
			ByteProcessor bp = new ByteProcessor(ip.getWidth(), ip.getHeight(), pixels8);
			bp.setColorModel(ip.getColorModel());
			return bp;
		}
		int[] pixels32 = (int[])ip.getPixels();
		double[] w = ColorProcessor.getWeightingFactors();
		if (((ColorProcessor)ip).getRGBWeights()!=null)
			w = ((ColorProcessor)ip).getRGBWeights();

Research_Associate · March 24, 2021, 9:30am

I think his original point was that this is incorrect (rounding differences) based on the results he listed. Thanks for linking the code.

The pixels are converted to grayscale using the formula
		g=r/3+g/3+b/3.

Which matches both posts above if the type for each operation is locked to an integer or byte, I suppose.

Bio7 · March 24, 2021, 9:30am

Thanks you @Research_Associate.

@Gwinzy: When I convert a created RGB image with your values red = 248, green = 177, blue = 149

I get a grayscale image with : 191

Which I also get with correct Java code.

Gwinzy · March 24, 2021, 9:41am

yes, you are right (MycroscopyRA). When i look at the pixel in imageJ they are the same as my code. If i save them as jpeg and look at them in matlab as i did the values are different, probably the converting to jpeg changed them slightly?
Thanks for your help everyone.

Research_Associate · March 24, 2021, 1:40pm

JPEG will change all of the values in your image - it is generally not considered appropriate for scientific analysis for this reason. Though, for whole slide images it is still frequently used due to file size.