2

I am trying to write an algorithm to rotate a bitmap image of $n$ by $n$ size by an angle $\alpha$.

I know that I have to find a rotation matrix, then perform matrix multiplication of the rotation matrix by the image matrix data input.

I know that the $2\times 2$ rotation matrix is

$$T_\alpha=\left(\begin{array}{cc}\cos\alpha & -\sin\alpha \\ \sin \alpha & \cos \alpha \end{array}\right)$$

However, I am not sure how to find the appropriate $n \times n$ matrix.

Alexander Gruber
  • 26,963
codigomonstruo
  • 121
  • 2
    @CheungJoonHee Not a duplicate. The O.P. isn’t looking for rotations in an $n$-dimensional vector space. The problem here is to rotate an $n\times n$ pixel array. – amd Apr 30 '19 at 04:57
  • 1
    You’re confusing the size of the image pixel array with the dimension of a vector space. You need an algorithm to rotate a pixel array. Try https://stackoverflow.com/q/484573 for suggestions. – amd Apr 30 '19 at 04:59
  • 1
    This is a 2-d rotation. You need to pick an origin and decide how to map the resulting rotated pixels into another $n \times n$ array. – copper.hat Apr 30 '19 at 05:00

1 Answers1

0

Let $w, h$ be the width & height of the image in pixels, but promote them to floats for further computation.

You take the image center $c = (w, h)/2$

Then treat each pixel coordinate as a coordinate with origin at $c$. In other words subtract $c$ from each pixel coordinate $v' = (x,y) - c$.

Now as you have the rotation matrix, you need only convert $v'$ to a column vector and multiply with $v'$ on the right.

If your rotation matrix was constructed using the right hand rule. And your image space (in whatever coding environment) originally started with $(0,0)$ at the bottom left, then this will rotate the image around its center $\alpha$ radians counter clockwise.

As far as final image quality though, this is probably bad once you discretize into a new image. This is why you should use something already coded.

Daniel Donnelly
  • 21,342