formula for distort function

[michaelba8]

Hi,
I'm implementing and testing a new algorithm for 3d points estimation, and right now I'm trying to test it on 3d virtual models.
I have downloaded and used your library for applying distortion on the images that I render from the 3d models.

In the test I render ideal images and apply the distortion on them, and I need to know how each ideal pixel is mapped to distorted pixel.
to get this information I need to know the math formula your library is using for each pixel, otherwise I have to apply distortion on each pixel alone to track the resulted pixel, and I have to this for many images, so the runtime will be very bad.

so if you can help with that and give me the actual implementation of the function:
def distort(self, method, arguments, best_fit=False)

or just the math formula that using the arguments list, any of the two will be perfect.

Thank you,
Michael

[emcconville]

The beauty of open source is that there's nothing stopping you from reading the source code. For distort formulas, check out distort.c in ImageMagick. The authors left plenty of math formulas as comments to help guide future readers.

Here's an example of a comment for PerspectiveDistortion

 /*
 Perspective Distortion (a ratio of affine distortions)
        p(x,y)    c0*x + c1*y + c2
    u = ------ = ------------------
        r(x,y)    c6*x + c7*y + 1
        q(x,y)    c3*x + c4*y + c5
    v = ------ = ------------------
        r(x,y)    c6*x + c7*y + 1
   c8 = Sign of 'r', or the denominator affine, for the actual image.
        This determines what part of the distorted image is 'ground'
        side of the horizon, the other part is 'sky' or invalid.
        Valid values are  +1.0  or  -1.0  only.
 Input Arguments are sets of control points...
 For Distort Images    u,v, x,y  ...
 For Sparse Gradients  x,y, r,g,b  ...
 Perspective Distortion Notes...
   + Can be thought of as ratio of  3 affine transformations
   + Not separatable: r() or c6 and c7 are used by both equations
   + All 8 coefficients must be determined simultaniously
   + Will only work with a 2 number_valuesal Image Distortion
   + Can not be used for generating a sparse gradient (interpolation)
   + It is not linear, but is simple to generate an inverse
   + All lines within an image remain lines.
   + but distances between points may vary.
*/

Most of the work in C is straight forward, but with a few look-up tables to manage coefficients & other repetitive routines.

[michaelba8]

Ohh I was looking for this source code, and couldn't find it.
Thank you very much for the fast reply, it was exactly what I needed.

[emcconville]

Awesome! Please mark this question as answered. It'll help future user searching for the same question. Also, Wand's distortion documents are in the process of being reworked. Feel free to comment on #534 if you feel there's something that needs extra attention.