Hello,

I am probably misunderstanding how the operators work and would like additional guidance. My requirement to set the individual Alpha channel per pixel based on the band[0] value, not set the band[0] channel to 0 or 255 based on the value.

Here is what is working with bitmap (DirectBitmap is a wrapper that allows direct manipulation of pixels "safely":

const byte grey = 220;
const byte min = 0;
const byte max = 255;
using (var srcImage = new Bitmap(origImage))
{
    using (var dstImage = new DirectBitmap(srcImage))
    {
        for (var i = 0; i < imageSize; i++)
        {
            for (var j = 0; j < imageSize; j++)
            {
                var shade = dstImage.GetR(j, i);
                var alpha = shade <= grey
                    ? max * (grey - shade) / (grey - min)
                    : max * (shade - grey) / (max - grey);
                dstImage.SetA(j, i, (byte)alpha);
            }
        }
    }
}

I need to be able to set a runtime calculated value for the alpha channel based on the current pixels Color.R value. Ifthenelse looks promising, but it seems to apply one of two preset masks instead of manipulating a value. Can I pass a function to an operator to achieve this?

Thanks much in advance!
-Timothy

You can just type your formula in. In python, perhaps:

#!/usr/bin/env python3

import sys
import pyvips

grey = 220
min = 0
max = 255

# we assume this is an RGB image
image = pyvips.Image.new_from_file(sys.argv[1], access="sequential")

shade = image[0]
alpha = (shade <= grey).ifthenelse(max * (grey - shade) / (grey - min),
                                   max * (shade - grey) / (max - grey))
# we add the alpha as an extra band to make an RGBA image
image = image.bandjoin(alpha.cast("uchar"))

image.write_to_file(sys.argv[2])

I see:

$ time ./alpha.py ~/pics/nina.jpg x.png

real	0m1.477s
user	0m1.796s
sys	0m0.160s

So 1.5s to make a 6k x 4k PNG:

Most of that time is in PNG write (rather a slow format!).

So the lines:

alpha = (shade <= grey).ifthenelse(max * (grey - shade) / (grey - min),
                                   max * (shade - grey) / (max - grey))

Are constructing two images:

then = max * (grey - shade) / (grey - min)
else = max * (shade - grey) / (max - grey)

and ifthenelse is picking between them using a third condition image generated by shade <= grey.

This sounds a little inefficient, but libvips is demand-driven, so none of these images really exist. Instead, each one is a scrap of code which can generate those pixels on demand --- the final write_to_file is pulling pixels through the system and evaluating these scraps of code in parallel and without ever constructing complete images.

For example, you can run the same python program on a large image:

$ /usr/bin/time -f %M:%e ./alpha.py ~/pics/st-francis.jpg x.v
665492:13.40

13s and 650mb of memory to make:

A 30,000 x 26,000 pixel image.

Ah OK. I would drop the alpha from your input image, otherwise you might have GAA (grey plus two alphas) which is very confusing.

shade = image[0]
alpha = ...
image = shade.bandjoin(alpha)

To make another GA image.