printMatrix, but no getMatrix

Programming Questions

BlueThen

printMatrix, but no getMatrix

in Programming Questions • 2 years ago

In summary: printMatrix() works fine with the P2D renderer, but I can't use getMatrix()

Hi. I'm trying to write a sprite class where I could load an image, and map screen coordinates to coordinates local to the sprite. For example, if I use push/pop matrix to rotate and translate the sprite to a certain part of a screen, and decide I want to draw a circle on the sprite with my mouse, I'll need to somehow take those coordinates, apply the rotations and translations that were applied to the sprite to those coordinates, and use them to edit the sprite's pixels array.

At the moment, I have a working version where I have a hashMap of screenX and screenY locations (with the index position in pixels[] for value) which I use to draw onto the sprite when needed. This hashMap gets updated every frame, or everytime the sprite changes positions. Unfortunately, this is too slow for me to put to good use.

So I'd like to do this a different way. Instead of storing the Sprite's real-world pixels in a hashMap, I'll just apply rotations/translations/scaling to the coordinates I'm using to edit the sprite. It'll be as simple as keeping track the applet's matrix at the time of the sprite's drawing, right?

I have one obstacle though. Processing (I'm using the P2D renderer) wont let me use the Matrix. Everytime I call getMatrix(), I get an error.

getMatrix(), or this particular variation of it, is not available with this renderer.

This confuses me though, because printMatrix() works just fine! Is this an error in P2D? Is it like this for a reason?

Perhaps I'm doing this completely wrong. I'm happy to take an advice regarding this. Thank you for your time.

Replies(4)

cedrickie..

Re: printMatrix, but no getMatrix

2 years ago

did you check out modelX,Y,Z ?
might be what you need if i understood you right.
http://processing.org/reference/modelX_.html

BlueThen

Re: Re: printMatrix, but no getMatrix

2 years ago

I mentioned that right now I'm using screenX,Y. However, I need the matrix values for my program to be efficient.

BlueThen

Re: printMatrix, but no getMatrix

2 years ago

I've been looking over Processing's source file for several hours, and I've FINALLY gotten a solution that seems to work.

It's very hacky, but I hope this can be helpful for anyone who'll need this in the future. (Can anyone add this to the wiki?)

import java.lang.reflect.Field; // you can place this at the beginning of the code for cleanliness
PMatrix2D hackMatrix () {
Class c = g.getClass();
PMatrix2D value = new PMatrix2D();
try {
// "Hack" into the PGraphics class to get ctm (the transformation matrix) and set it to public, instead of private.
Field field = c.getDeclaredField("ctm");
field.setAccessible(true);
try {
value = (PMatrix2D) field.get(g);
}
catch (IllegalAccessException e) {
throw new RuntimeException("Cannot attain matrix");
}
}
catch (NoSuchFieldException e) {
throw new RuntimeException("Didn't find field named '");
}
return value;
}

Also, I'd be cautious because I'm sure that getMatrix is disabled for a reason, so it might not always work.

BlueThen

Re: printMatrix, but no getMatrix

2 years ago

Hi again. My next issue was reversing the transformation matrix. Processing uses matrix mulitiplication when applying a transformation matrix to a 2D coordinate.

// taken from Processing's PGraphics2D source.
// note that ctm is a matrix, and m## are numbers within the matrix
public float screenX(float x, float y) {
return ctm.m00 * x + ctm.m01 * y + ctm.m02;
}
public float screenY(float x, float y) {
return ctm.m10 * x + ctm.m11 * y + ctm.m12
}

Normally, we could just invert the matrix and multiply it against x and y, however the matrix is invertible due to the determinant.

Instead, we'll have to use some algebra and solve for x and y.

x' = ctm.m00 * x + ctm.m01 * y + ctm.m02;
y' = ctm.m10 * x + ctm.m11 * y + ctm.m12
// Magical math stuff is done here
// I cheated and used WolframAlpha.
x = (-ctm.m01 * (ctm.m12 - y') + ctm.m11 * (ctm.m02 - x')) / ((ctm.m01 * ctm.m10) - (ctm.m00 * ctm.m11))
y = ( ctm.m00 * (ctm.m12 - y') - ctm.m10 * (ctm.m02 - x')) / ((ctm.m01 * ctm.m10) - (ctm.m00 * ctm.m11))

After optimizing for a bit, this is the function I came up with:

// takes a transformation matrix, an x and y coordinate with the matrix already applied to it, and reverses the transformations
// returns an integer array with return[0] being x, and return[1] being y
int[] undoMatrix (PMatrix2D matrix, float x, float y) {
// to minimize the amount of calculations we need, we store some of the math inside variables
double m01_x_m10__Minus__m00_x_m11 = (matrix.m01 * matrix.m10) - (matrix.m00 * matrix.m11);
double m02minusX = matrix.m02 - x;
double m12minusY = matrix.m12 - y;
// solve for x and y (Thanks to http://www.wolframalpha.com/input/?i=Reduce[{g+==+a+x+++b+y+++c,+h+=+d+x+++e+y+++f},+{x,y}] )
int newX = (int) ((-matrix.m01 * m12minusY + matrix.m11 * m02minusX) / (m01_x_m10__Minus__m00_x_m11));
int newY = (int) (( matrix.m00 * m12minusY - matrix.m10 * m02minusX) / (m01_x_m10__Minus__m00_x_m11));
return new int[]{newX, newY};
}

I hope this helps someone out there. I've had issues with this kind of thing several times in the past. I'm still

unsure why Processing's dev team won't allow us to use getMatrix, unless it's because invert() doesn't work.

Edit: I submitted an issue: http://code.google.com/p/processing/issues/detail?id=527

Edit2: Seems that this code isn't completely accurate. I'm looking into it.

Edit3: It wasn't this code. It was an error in another part of the program on my part.

Top Reply