In-memory bitmap transformation

Question

I have the following problem. I started with a bitmap transformation routine that works flawlessly for any kind of transformation I can throw at it.

  Bitmap transform(Bitmap src) {
    // ... any kind of transformation , for example GAMMA
    double gama = 0.8;
    int[] tR = new int[256];
    int[] gG = new int[256];
    int[] tB = new int[256];
    for(int i = 0; i < 256; ++i) {
      tR[i] = (int)Math.min(255, (int)((255.0 * Math.pow(i / 255.0, 1.0 / gama)) + 0.5));
      tG[i] = (int)Math.min(255, (int)((255.0 * Math.pow(i / 255.0, 1.0 / gama)) + 0.5));
      tB[i] = (int)Math.min(255, (int)((255.0 * Math.pow(i / 255.0, 1.0 / gama)) + 0.5));
    }
    // apply transformation to the old bitmap -> bmOut  
    int wid = src.getWidth(), hei = src.getHeight();
    Bitmap bmOut = Bitmap.createBitmap(wid, hei, src.getConfig());
    int A, R, G, B;
    for(int x = 0; x < wid; x++) {
      for(int y = 0; y < hei; y++) {
        int pixel = src.getPixel(x, y);
        A = Color.alpha(pixel);
        R = tR[Color.red(pixel)];
        G = tG[Color.green(pixel)];
        B = tB[Color.blue(pixel)];
        bmOut.setPixel(x, y, Color.argb(A, R, G, B));
      }
    }
    return bmOut;
  }

But it is PAINFULLY slow - caused by the getPixel() / setPixel() brothers, sisters. No problem, says I, I'll just use a memory buffer (like in the old StretchBlt() days). So, I did a MAJOR rewrite, creating the following gem of software engineering :)

  Bitmap transform(Bitmap src) {
    // ... transformation array are built here

    // apply transformation  
    int wid = src.getWidth(), hei = src.getHeight();
    Bitmap bmOut = Bitmap.createBitmap(wid, hei, src.getConfig());

    int[] pixs = new int[wid*hei];                  // changed
    src.getPixels(pixs, 0, wid, 0, 0, wid, hei);    // changed

    int A, R, G, B;
    for(int x = 0; x < wid; x++) {
      for(int y = 0; y < hei; y++) {
        int off = ( x * y ) + y;                    // changed
        int pixel = pixs[off];                      // changed
        A = Color.alpha(pixel);
        R = tR[Color.red(pixel)];
        G = tG[Color.green(pixel)];
        B = tB[Color.blue(pixel)];
        pixs[off] = Color.argb(A, R, G, B);         // changed      
      }
    }
    bmOut.setPixels(pixs, 0, wid, 0, 0, wid, hei);  // changed
    return bmOut;
  }

Runs fast, even gets a correct result IF THERE IS NO TRANSFORMATION. But it falls apart if I try to massage the pixels (apply transformations). So I did comparison of ARGB pixels from getPixel() vs array of pixels values from getPixels(...) and they are different (well, the first 2 are the same, which leaves me with about zillion that are not).

array     getPixel
a r g b   a r g b
------------------
ff65340b  ff65340b
ff64330a  ff64330a
ff66320b  ff63320a
ff65310a  ff613008
ff66300c  ff62300d
ff67310d  ff62300d
ff68300d  ff622d0d
ff69310e  ff5f2a0a
....

Anybody knows what I'm doing wrong this time? I am not willing to give up the speed of the mem-array solution yet. Thanks, sean

Answer 1

It should be

int off = ( y * wid ) + x;

By the way, I think the two loops is unnecessary, you can simply do:

for (int off = pixs.length - 1; off >= 0; off--)