Alpha blending sometimes gives the wrong color?

Question

I've got a method for alpha blending:

public static int alphaBlend(int c1, int c2) {
        int a1 = (c1 & 0xff000000) >>> 24;

        int r1 = (c1 & 0x00ff0000) >> 16;
        int r2 = (c2 & 0x00ff0000) >> 16;

        int g1 = (c1 & 0x0000ff00) >> 8;
        int g2 = (c2 & 0x0000ff00) >> 8;

        int b1 = (c1 & 0x000000ff);
        int b2 = (c2 & 0x000000ff);

        float src_alpha = ((float)a1) / 255.0f;

        int red   = (int) ((r1 * src_alpha) + (r2 * (1.0f - src_alpha)));
        int green = (int) ((g1 * src_alpha) + (g2 * (1.0f - src_alpha)));
        int blue  = (int) ((b1 * src_alpha) + (b2 * (1.0f - src_alpha)));

        return (a1 << 24) | (red << 16) | (green << 8) | blue;
    }

But sometimes this gives the wrong color (I guess because I don't use the second color's alpha). Could anyone explain how to fix this?

Answer 1

What do you mean by "wrong"?

Your first step could be to say clearly (to yourself, or in a unit test, or in this question) "I expected alphaBlend(something,something) to produce (something, preferably a number), but what I got was (something else)." Like, blended red and blue, expected purple, got mauve. Or whatever. Hex values would be clearer...

At a quick reading of the arithmetic, the color output looks correct, although the resulting alpha for a "source over" compositing operation is usually:

a-result = a1 + a2 - a1 * a2

http://en.wikipedia.org/wiki/Alpha_compositing#Alpha_blending