I have a 2d matrix created from position, scale and rotation (no skew). I would like to be able to decompose this matrix back to the original components and have managed to do so with the following pseudo code:
posX = matrix.tx
posY = matrix.ty
scaleX = Sqrt( matrix.a * matrix.a + matrix.b * matrix.b )
scaleY = Sqrt( matrix.c * matrix.c + matrix.d * matrix.d )
rotation = ATan2( -matrix.c / scaleY, matrix.a / scaleX )
However this obviously only works with positive scale values and I am unsure how to calculate the correct negative scales. I have attempted various suggestions found using google but so far none have worked correctly.
I have tried the accepted answer from here and the decomposition explained here, whilst they produce correct transformations, the components of scale and rotation do not match my original values.
I have tried taking the sign of the diagonal matrix.a * matrix.d which appears to work for the scale on the x axis but unsure if this is the correct approach and can't figure out how to handle the y axis.
Is this even possible? Will I have to accept that I will not get back the exact components and the best I can hope for is values that produce the same transformation?
Any help or pointers would be greatly appreciated.
Original
Translation = 204, 159
Rotation = -3.0168146900000044
Scale = -3, -2
Matrix = [ 2.976675975304773, 0.37336327891663146, -0.24890885261108764, 1.984450650203182, 204, 159 ]
Decomposition
Translation = 204, 159
Rotation = 0.1247779635897889
Scale = 3, 2
Matrix = [ 2.976675975304773, 0.3733632789166315, -0.24890885261108767, 1.984450650203182, 204, 159 ]
That was using the following decomposition code:
posX = matrix.tx
posY = matrix.ty
scaleX = Sgn( a ) * Sqrt( matrix.a * matrix.a + matrix.b * matrix.b )
scaleY = Sgn( d ) * Sqrt( matrix.c * matrix.c + matrix.d * matrix.d )
rotation = ATan2( -matrix.c / scaleY, matrix.a / scaleX )
Sometimes you can't tell flip (negative scale) from rotation, e.g. an image flipped horizontally and vertically is identical to the image rotated by 180 degrees.
So what you have to do is know whether the transformation matrix contains flips. With that knowledge, you can cancel out the flip first, decompose it as usual, and put the flip back into the decomposed scale factors.
Pseudo code:
Matrix m
hFlip = true
vFlip = true
if hFlip: m = compose(m, scale(-1, 1))
if vFlip: m = compose(m, scale(1, -1))
translation, rotation, scale = decompose(m)
if hFlip: scale.x = -scale.x
if vFlip: scale.y = -scale.y
Related
I think these should be circular. I assume there is something wrong with my normals but I haven't found anything wrong with them. Then again, finding a good test for the normals is difficult.
Here is the image:
Here is my shading code for each light, leaving out the recursive part for reflections:
lighting = ( hit.obj.ambient + hit.obj.emission );
const glm::vec3 view_direction = glm::normalize(eye - hit.pos);
const glm::vec3 reflection = glm::normalize(( static_cast<float>(2) * ( glm::dot(view_direction, hit.normal) * hit.normal ) ) - view_direction);
for(int i = 0; i < numused; ++i)
{
glm::vec3 hit_to_light = (lights[i].pos - hit.pos);
float dist = glm::length(hit_to_light);
glm::vec3 light_direction = glm::normalize(hit_to_light);
Ray lightray(hit.pos, light_direction);
Intersection blocked = Intersect(lightray, scene, verbose ? verbose : false);
if( blocked.dist >= dist)
{
glm::vec3 halfangle = glm::normalize(view_direction + light_direction);
float specular_multiplier = pow(std::max(glm::dot(halfangle,hit.normal), 0.f), shininess);
glm::vec3 attenuation_term = lights[i].rgb * (1.0f / (attenuation + dist * linear + dist*dist * quad));
glm::vec3 diffuse_term = hit.obj.diffuse * ( std::max(glm::dot(light_direction,hit.normal) , 0.f) );
glm::vec3 specular_term = hit.obj.specular * specular_multiplier;
}
}
And here is the line where I transform the object space normal to world space:
*norm = glm::normalize(transinv * glm::vec4(glm::normalize(p - sphere_center), 0));
Using the full phong model, instead of blinn-phong, I get teardrop highlights:
If I color pixels according to the (absolute value of the) normal at the intersection point I get the following image (r = x, g = y, b = z):
I've solved this issue. It turns out that the normals were all just slightly off, but not enough that the image colored by normals could depict it.
I found this out by computing the normals on spheres with a uniform scale and a translation.
The problem occurred in the line where I transformed the normals to world space:
*norm = glm::normalize(transinv * glm::vec4(glm::normalize(p - sphere_center), 0));
I assumed that the homogeneous coordinate would be 0 after the transformation because it was zero beforehand (rotations and scales do not affect it, and because it is 0, neither can translations). However, it is not 0 because the matrix is transposed, so the bottom row was filled with the inverse translations, causing the homogeneous coordinate to be nonzero.
The 4-vector is then normalized and the result is assigned to a 3-vector. The constructor for the 3-vector simply removes the last entry, so the normal was left unnormalized.
Here's the final picture:
BOUNTY STATUS UPDATE:
I discovered how to map a linear lens, from destination coordinates to source coordinates.
How do you calculate the radial distance from the centre to go from fisheye to rectilinear?
1). I actually struggle to reverse it, and to map source coordinates to destination coordinates. What is the inverse, in code in the style of the converting functions I posted?
2). I also see that my undistortion is imperfect on some lenses - presumably those that are not strictly linear. What is the equivalent to-and-from source-and-destination coordinates for those lenses? Again, more code than just mathematical formulae please...
Question as originally stated:
I have some points that describe positions in a picture taken with a fisheye lens.
I want to convert these points to rectilinear coordinates. I want to undistort the image.
I've found this description of how to generate a fisheye effect, but not how to reverse it.
There's also a blog post that describes how to use tools to do it; these pictures are from that:
(1) : SOURCE Original photo link
Input : Original image with fish-eye distortion to fix.
(2) : DESTINATION Original photo link
Output : Corrected image (technically also with perspective correction, but that's a separate step).
How do you calculate the radial distance from the centre to go from fisheye to rectilinear?
My function stub looks like this:
Point correct_fisheye(const Point& p,const Size& img) {
// to polar
const Point centre = {img.width/2,img.height/2};
const Point rel = {p.x-centre.x,p.y-centre.y};
const double theta = atan2(rel.y,rel.x);
double R = sqrt((rel.x*rel.x)+(rel.y*rel.y));
// fisheye undistortion in here please
//... change R ...
// back to rectangular
const Point ret = Point(centre.x+R*cos(theta),centre.y+R*sin(theta));
fprintf(stderr,"(%d,%d) in (%d,%d) = %f,%f = (%d,%d)\n",p.x,p.y,img.width,img.height,theta,R,ret.x,ret.y);
return ret;
}
Alternatively, I could somehow convert the image from fisheye to rectilinear before finding the points, but I'm completely befuddled by the OpenCV documentation. Is there a straightforward way to do it in OpenCV, and does it perform well enough to do it to a live video feed?
The description you mention states that the projection by a pin-hole camera (one that does not introduce lens distortion) is modeled by
R_u = f*tan(theta)
and the projection by common fisheye lens cameras (that is, distorted) is modeled by
R_d = 2*f*sin(theta/2)
You already know R_d and theta and if you knew the camera's focal length (represented by f) then correcting the image would amount to computing R_u in terms of R_d and theta. In other words,
R_u = f*tan(2*asin(R_d/(2*f)))
is the formula you're looking for. Estimating the focal length f can be solved by calibrating the camera or other means such as letting the user provide feedback on how well the image is corrected or using knowledge from the original scene.
In order to solve the same problem using OpenCV, you would have to obtain the camera's intrinsic parameters and lens distortion coefficients. See, for example, Chapter 11 of Learning OpenCV (don't forget to check the correction). Then you can use a program such as this one (written with the Python bindings for OpenCV) in order to reverse lens distortion:
#!/usr/bin/python
# ./undistort 0_0000.jpg 1367.451167 1367.451167 0 0 -0.246065 0.193617 -0.002004 -0.002056
import sys
import cv
def main(argv):
if len(argv) < 10:
print 'Usage: %s input-file fx fy cx cy k1 k2 p1 p2 output-file' % argv[0]
sys.exit(-1)
src = argv[1]
fx, fy, cx, cy, k1, k2, p1, p2, output = argv[2:]
intrinsics = cv.CreateMat(3, 3, cv.CV_64FC1)
cv.Zero(intrinsics)
intrinsics[0, 0] = float(fx)
intrinsics[1, 1] = float(fy)
intrinsics[2, 2] = 1.0
intrinsics[0, 2] = float(cx)
intrinsics[1, 2] = float(cy)
dist_coeffs = cv.CreateMat(1, 4, cv.CV_64FC1)
cv.Zero(dist_coeffs)
dist_coeffs[0, 0] = float(k1)
dist_coeffs[0, 1] = float(k2)
dist_coeffs[0, 2] = float(p1)
dist_coeffs[0, 3] = float(p2)
src = cv.LoadImage(src)
dst = cv.CreateImage(cv.GetSize(src), src.depth, src.nChannels)
mapx = cv.CreateImage(cv.GetSize(src), cv.IPL_DEPTH_32F, 1)
mapy = cv.CreateImage(cv.GetSize(src), cv.IPL_DEPTH_32F, 1)
cv.InitUndistortMap(intrinsics, dist_coeffs, mapx, mapy)
cv.Remap(src, dst, mapx, mapy, cv.CV_INTER_LINEAR + cv.CV_WARP_FILL_OUTLIERS, cv.ScalarAll(0))
# cv.Undistort2(src, dst, intrinsics, dist_coeffs)
cv.SaveImage(output, dst)
if __name__ == '__main__':
main(sys.argv)
Also note that OpenCV uses a very different lens distortion model to the one in the web page you linked to.
(Original poster, providing an alternative)
The following function maps destination (rectilinear) coordinates to source (fisheye-distorted) coordinates. (I'd appreciate help in reversing it)
I got to this point through trial-and-error: I don't fundamentally grasp why this code is working, explanations and improved accuracy appreciated!
def dist(x,y):
return sqrt(x*x+y*y)
def correct_fisheye(src_size,dest_size,dx,dy,factor):
""" returns a tuple of source coordinates (sx,sy)
(note: values can be out of range)"""
# convert dx,dy to relative coordinates
rx, ry = dx-(dest_size[0]/2), dy-(dest_size[1]/2)
# calc theta
r = dist(rx,ry)/(dist(src_size[0],src_size[1])/factor)
if 0==r:
theta = 1.0
else:
theta = atan(r)/r
# back to absolute coordinates
sx, sy = (src_size[0]/2)+theta*rx, (src_size[1]/2)+theta*ry
# done
return (int(round(sx)),int(round(sy)))
When used with a factor of 3.0, it successfully undistorts the images used as examples (I made no attempt at quality interpolation):
Dead link
(And this is from the blog post, for comparison:)
If you think your formulas are exact, you can comput an exact formula with trig, like so:
Rin = 2 f sin(w/2) -> sin(w/2)= Rin/2f
Rout= f tan(w) -> tan(w)= Rout/f
(Rin/2f)^2 = [sin(w/2)]^2 = (1 - cos(w))/2 -> cos(w) = 1 - 2(Rin/2f)^2
(Rout/f)^2 = [tan(w)]^2 = 1/[cos(w)]^2 - 1
-> (Rout/f)^2 = 1/(1-2[Rin/2f]^2)^2 - 1
However, as #jmbr says, the actual camera distortion will depend on the lens and the zoom. Rather than rely on a fixed formula, you might want to try a polynomial expansion:
Rout = Rin*(1 + A*Rin^2 + B*Rin^4 + ...)
By tweaking first A, then higher-order coefficients, you can compute any reasonable local function (the form of the expansion takes advantage of the symmetry of the problem). In particular, it should be possible to compute initial coefficients to approximate the theoretical function above.
Also, for good results, you will need to use an interpolation filter to generate your corrected image. As long as the distortion is not too great, you can use the kind of filter you would use to rescale the image linearly without much problem.
Edit: as per your request, the equivalent scaling factor for the above formula:
(Rout/f)^2 = 1/(1-2[Rin/2f]^2)^2 - 1
-> Rout/f = [Rin/f] * sqrt(1-[Rin/f]^2/4)/(1-[Rin/f]^2/2)
If you plot the above formula alongside tan(Rin/f), you can see that they are very similar in shape. Basically, distortion from the tangent becomes severe before sin(w) becomes much different from w.
The inverse formula should be something like:
Rin/f = [Rout/f] / sqrt( sqrt(([Rout/f]^2+1) * (sqrt([Rout/f]^2+1) + 1) / 2 )
I blindly implemented the formulas from here, so I cannot guarantee it would do what you need.
Use auto_zoom to get the value for the zoom parameter.
def dist(x,y):
return sqrt(x*x+y*y)
def fisheye_to_rectilinear(src_size,dest_size,sx,sy,crop_factor,zoom):
""" returns a tuple of dest coordinates (dx,dy)
(note: values can be out of range)
crop_factor is ratio of sphere diameter to diagonal of the source image"""
# convert sx,sy to relative coordinates
rx, ry = sx-(src_size[0]/2), sy-(src_size[1]/2)
r = dist(rx,ry)
# focal distance = radius of the sphere
pi = 3.1415926535
f = dist(src_size[0],src_size[1])*factor/pi
# calc theta 1) linear mapping (older Nikon)
theta = r / f
# calc theta 2) nonlinear mapping
# theta = asin ( r / ( 2 * f ) ) * 2
# calc new radius
nr = tan(theta) * zoom
# back to absolute coordinates
dx, dy = (dest_size[0]/2)+rx/r*nr, (dest_size[1]/2)+ry/r*nr
# done
return (int(round(dx)),int(round(dy)))
def fisheye_auto_zoom(src_size,dest_size,crop_factor):
""" calculate zoom such that left edge of source image matches left edge of dest image """
# Try to see what happens with zoom=1
dx, dy = fisheye_to_rectilinear(src_size, dest_size, 0, src_size[1]/2, crop_factor, 1)
# Calculate zoom so the result is what we wanted
obtained_r = dest_size[0]/2 - dx
required_r = dest_size[0]/2
zoom = required_r / obtained_r
return zoom
I took what JMBR did and basically reversed it. He took the radius of the distorted image (Rd, that is, the distance in pixels from the center of the image) and found a formula for Ru, the radius of the undistorted image.
You want to go the other way. For each pixel in the undistorted (processed image), you want to know what the corresponding pixel is in the distorted image.
In other words, given (xu, yu) --> (xd, yd). You then replace each pixel in the undistorted image with its corresponding pixel from the distorted image.
Starting where JMBR did, I do the reverse, finding Rd as a function of Ru. I get:
Rd = f * sqrt(2) * sqrt( 1 - 1/sqrt(r^2 +1))
where f is the focal length in pixels (I'll explain later), and r = Ru/f.
The focal length for my camera was 2.5 mm. The size of each pixel on my CCD was 6 um square. f was therefore 2500/6 = 417 pixels. This can be found by trial and error.
Finding Rd allows you to find the corresponding pixel in the distorted image using polar coordinates.
The angle of each pixel from the center point is the same:
theta = arctan( (yu-yc)/(xu-xc) ) where xc, yc are the center points.
Then,
xd = Rd * cos(theta) + xc
yd = Rd * sin(theta) + yc
Make sure you know which quadrant you are in.
Here is the C# code I used
public class Analyzer
{
private ArrayList mFisheyeCorrect;
private int mFELimit = 1500;
private double mScaleFESize = 0.9;
public Analyzer()
{
//A lookup table so we don't have to calculate Rdistorted over and over
//The values will be multiplied by focal length in pixels to
//get the Rdistorted
mFisheyeCorrect = new ArrayList(mFELimit);
//i corresponds to Rundist/focalLengthInPixels * 1000 (to get integers)
for (int i = 0; i < mFELimit; i++)
{
double result = Math.Sqrt(1 - 1 / Math.Sqrt(1.0 + (double)i * i / 1000000.0)) * 1.4142136;
mFisheyeCorrect.Add(result);
}
}
public Bitmap RemoveFisheye(ref Bitmap aImage, double aFocalLinPixels)
{
Bitmap correctedImage = new Bitmap(aImage.Width, aImage.Height);
//The center points of the image
double xc = aImage.Width / 2.0;
double yc = aImage.Height / 2.0;
Boolean xpos, ypos;
//Move through the pixels in the corrected image;
//set to corresponding pixels in distorted image
for (int i = 0; i < correctedImage.Width; i++)
{
for (int j = 0; j < correctedImage.Height; j++)
{
//which quadrant are we in?
xpos = i > xc;
ypos = j > yc;
//Find the distance from the center
double xdif = i-xc;
double ydif = j-yc;
//The distance squared
double Rusquare = xdif * xdif + ydif * ydif;
//the angle from the center
double theta = Math.Atan2(ydif, xdif);
//find index for lookup table
int index = (int)(Math.Sqrt(Rusquare) / aFocalLinPixels * 1000);
if (index >= mFELimit) index = mFELimit - 1;
//calculated Rdistorted
double Rd = aFocalLinPixels * (double)mFisheyeCorrect[index]
/mScaleFESize;
//calculate x and y distances
double xdelta = Math.Abs(Rd*Math.Cos(theta));
double ydelta = Math.Abs(Rd * Math.Sin(theta));
//convert to pixel coordinates
int xd = (int)(xc + (xpos ? xdelta : -xdelta));
int yd = (int)(yc + (ypos ? ydelta : -ydelta));
xd = Math.Max(0, Math.Min(xd, aImage.Width-1));
yd = Math.Max(0, Math.Min(yd, aImage.Height-1));
//set the corrected pixel value from the distorted image
correctedImage.SetPixel(i, j, aImage.GetPixel(xd, yd));
}
}
return correctedImage;
}
}
I found this pdf file and I have proved that the maths are correct (except for the line vd = *xd**fv+v0 which should say vd = **yd**+fv+v0).
http://perception.inrialpes.fr/CAVA_Dataset/Site/files/Calibration_OpenCV.pdf
It does not use all of the latest co-efficients that OpenCV has available but I am sure that it could be adapted fairly easily.
double k1 = cameraIntrinsic.distortion[0];
double k2 = cameraIntrinsic.distortion[1];
double p1 = cameraIntrinsic.distortion[2];
double p2 = cameraIntrinsic.distortion[3];
double k3 = cameraIntrinsic.distortion[4];
double fu = cameraIntrinsic.focalLength[0];
double fv = cameraIntrinsic.focalLength[1];
double u0 = cameraIntrinsic.principalPoint[0];
double v0 = cameraIntrinsic.principalPoint[1];
double u, v;
u = thisPoint->x; // the undistorted point
v = thisPoint->y;
double x = ( u - u0 )/fu;
double y = ( v - v0 )/fv;
double r2 = (x*x) + (y*y);
double r4 = r2*r2;
double cDist = 1 + (k1*r2) + (k2*r4);
double xr = x*cDist;
double yr = y*cDist;
double a1 = 2*x*y;
double a2 = r2 + (2*(x*x));
double a3 = r2 + (2*(y*y));
double dx = (a1*p1) + (a2*p2);
double dy = (a3*p1) + (a1*p2);
double xd = xr + dx;
double yd = yr + dy;
double ud = (xd*fu) + u0;
double vd = (yd*fv) + v0;
thisPoint->x = ud; // the distorted point
thisPoint->y = vd;
This can be solved as an optimization problem. Simply draw on curves in images that are supposed to be straight lines. Store the contour points for each of those curves. Now we can solve the fish eye matrix as a minimization problem. Minimize the curve in points and that will give us a fisheye matrix. It works.
It can be done manually by adjusting the fish eye matrix using trackbars! Here is a fish eye GUI code using OpenCV for manual calibration.
I'm using jsc3d to load and display some 3d objects on a canvas. The viewer has already a built-in feature that allows to rotate the "view coordinates" (correct me if i'm wrong) about the Y axis by dragging the mouse.
The rotation is performed through a classic rotation matrix, and finally the trasformation matrix is multiplied by this rotation matrix.
The totation about the Y axis is calculated in a way that resembles a circular movement around the whole scene of loaded objects:
JSC3D.Matrix3x4.prototype.rotateAboutYAxis = function(angle) {
if(angle != 0) {
angle *= Math.PI / 180;
var c = Math.cos(angle);
var s = Math.sin(angle);
var m00 = c * this.m00 + s * this.m20;
var m01 = c * this.m01 + s * this.m21;
var m02 = c * this.m02 + s * this.m22;
var m03 = c * this.m03 + s * this.m23;
var m20 = c * this.m20 - s * this.m00;
var m21 = c * this.m21 - s * this.m01;
var m22 = c * this.m22 - s * this.m02;
var m23 = c * this.m23 - s * this.m03;
this.m00 = m00; this.m01 = m01; this.m02 = m02; this.m03 = m03;
this.m20 = m20; this.m21 = m21; this.m22 = m22; this.m23 = m23;
}
};
Now, dragging the mouse will apply this rotation about the Y axis on the whole world, like on the left side in the picture below. Is there a way, to apply a rotation about the Up vector to keep it in the initial position, like it appear on the right side?
I tried something like that:
var rotY = (x - viewer.mouseX) * 360 / viewer.canvas.height;
var rotMat = new JSC3D.Matrix3x4; // identity
rotMat.rotateAboutYAxis(rotY);
viewer.rotMatrix.multiply(rotMat);
but it has no effect.
What operations shall be applied to my rotation matrix to achieve a rotation about the Up vector?
Sample: https://jsfiddle.net/4xzjnnar/1/
This 3D library has already some built-in functions to allow scene rotation about X,Y,and Z axis, so there is no need to implement new matrix operations for that, we can use the existing functions rotateAboutXAyis, rotateAboutYAxis and rotateAboutZAxis, which apply an in-place matrix multiplication of the desired rotation angle in degrees.
The scene in JSC3D is transformed by a 3x4 matrix where the rotation is stored in the first 3 values of each row.
After applying a scene rotation and/or translation, applying a subsequent rotation about the Up vector, is a problem of calculate a rotation about an arbitrary axis.
A very clean and didactic explanation how to solve this problem is described here: http://ami.ektf.hu/uploads/papers/finalpdf/AMI_40_from175to186.pdf
Translate the P 0 (x 0 ,y 0 ,z 0 ) axis point to the origin of the coordinate system.
Perform appropriate rotations to make the axis of rotation coincident with
z-coordinate axis.
Rotate about the z-axis by the angle θ.
Perform the inverse of the combined rotation transformation.
Perform the inverse of the translation.
Now, its easy to write a function for that, because we use the functions already available in JSC3D (translation part is omitted here).
JSC3D.Viewer.prototype.rotateAboutUpVector = function(angle) {
angle %= 360;
/* pitch, counter-clockwise rotation about the Y axis */
var degX = this.rpy[0], degZ = this.rpy[2];
this.rotMatrix.rotateAboutXAxis(-degX);
this.rotMatrix.rotateAboutZAxis(-degZ);
this.rotMatrix.rotateAboutYAxis(angle);
this.rotMatrix.rotateAboutZAxis(degZ);
this.rotMatrix.rotateAboutXAxis(degX);
}
Because all above mentioned functions are using degrees, we need to get back the actual Euler angles from the rotation matrix (simplified):
JSC3D.Viewer.prototype.calcRollPitchYaw = function() {
var m = this.rotMatrix;
var radians = 180 / Math.PI;
var angleX = Math.atan2(-m.m12, m.m22) * radians;
var angleY = Math.asin(m.m01) * radians;
var angleZ = Math.atan2(-m.m01, m.m00) * radians;
this.rpy[0] = angleX;
this.rpy[1] = angleY;
this.rpy[2] = angleZ;
}
The tricky part here, is that we need always to get back the current rotation angles, as they results from the applied rotations, so a separate function must be used to store the current Euler angles every time that a rotation is applied to the scene.
For that, we can use a very simple structure:
JSC3D.Viewer.prototype.rpy = [0, 0, 0];
This will be the final result:
i have two rotation matrices that describe arbitrary rotations. (4x4 opengl compatible)
now i want to interpolate between them, so that it follows a radial path from one rotation to the other. think of a camera on a tripod looking one way and then rotating.
if i interpolate every component i get a squeezing result, so i think i need to interpolate only certain components of the matrix. but which ones?
You have to use SLERP for the rotational parts of the matrices, and linear for the other parts. The best way is to turn your matrices into quaternions and use the (simpler) quaternion SLERP: http://en.wikipedia.org/wiki/Slerp.
I suggest reading Graphic Gems II or III,specifically the sections about decomposing matrices into simpler transformations. Here's Spencer W. Thomas' source for this chapter:
http://tog.acm.org/resources/GraphicsGems/gemsii/unmatrix.c
Of course, I suggest you learn how to do this yourself. It's really not that hard, just a lot of annoying algebra. And finally, here's a great paper on how to turn a matrix into a quaternion, and back, by Id software: http://www.mrelusive.com/publications/papers/SIMD-From-Quaternion-to-Matrix-and-Back.pdf
Edit: This is the formula pretty much everyone cites, it's from a 1985 SIGGRAPH paper.
Where:
- qm = interpolated quaternion
- qa = quaternion a (first quaternion to be interpolated between)
- qb = quaternion b (second quaternion to be interpolated between)
- t = a scalar between 0.0 (at qa) and 1.0 (at qb)
- θ is half the angle between qa and qb
Code:
quat slerp(quat qa, quat qb, double t) {
// quaternion to return
quat qm = new quat();
// Calculate angle between them.
double cosHalfTheta = qa.w * qb.w + qa.x * qb.x + qa.y * qb.y + qa.z * qb.z;
// if qa=qb or qa=-qb then theta = 0 and we can return qa
if (abs(cosHalfTheta) >= 1.0){
qm.w = qa.w;qm.x = qa.x;qm.y = qa.y;qm.z = qa.z;
return qm;
}
// Calculate temporary values.
double halfTheta = acos(cosHalfTheta);
double sinHalfTheta = sqrt(1.0 - cosHalfTheta*cosHalfTheta);
// if theta = 180 degrees then result is not fully defined
// we could rotate around any axis normal to qa or qb
if (fabs(sinHalfTheta) < 0.001){ // fabs is floating point absolute
qm.w = (qa.w * 0.5 + qb.w * 0.5);
qm.x = (qa.x * 0.5 + qb.x * 0.5);
qm.y = (qa.y * 0.5 + qb.y * 0.5);
qm.z = (qa.z * 0.5 + qb.z * 0.5);
return qm;
}
double ratioA = sin((1 - t) * halfTheta) / sinHalfTheta;
double ratioB = sin(t * halfTheta) / sinHalfTheta;
//calculate Quaternion.
qm.w = (qa.w * ratioA + qb.w * ratioB);
qm.x = (qa.x * ratioA + qb.x * ratioB);
qm.y = (qa.y * ratioA + qb.y * ratioB);
qm.z = (qa.z * ratioA + qb.z * ratioB);
return qm;
}
From: http://www.euclideanspace.com/maths/algebra/realNormedAlgebra/quaternions/slerp/
You need to convert the matrix into a different representation - quaternions work well for this, and interpolating quaternions is a well-defined operation.
I'm currently writing a program to generate really enormous (65536x65536 pixels and above) Mandelbrot images, and I'd like to devise a spectrum and coloring scheme that does them justice. The wikipedia featured mandelbrot image seems like an excellent example, especially how the palette remains varied at all zoom levels of the sequence. I'm not sure if it's rotating the palette or doing some other trick to achieve this, though.
I'm familiar with the smooth coloring algorithm for the mandelbrot set, so I can avoid banding, but I still need a way to assign colors to output values from this algorithm.
The images I'm generating are pyramidal (eg, a series of images, each of which has half the dimensions of the previous one), so I can use a rotating palette of some sort, as long as the change in the palette between subsequent zoom levels isn't too obvious.
This is the smooth color algorithm:
Lets say you start with the complex number z0 and iterate n times until it escapes. Let the end point be zn.
A smooth value would be
nsmooth := n + 1 - Math.log(Math.log(zn.abs()))/Math.log(2)
This only works for mandelbrot, if you want to compute a smooth function for julia sets, then use
Complex z = new Complex(x,y);
double smoothcolor = Math.exp(-z.abs());
for(i=0;i<max_iter && z.abs() < 30;i++) {
z = f(z);
smoothcolor += Math.exp(-z.abs());
}
Then smoothcolor is in the interval (0,max_iter).
Divide smoothcolor with max_iter to get a value between 0 and 1.
To get a smooth color from the value:
This can be called, for example (in Java):
Color.HSBtoRGB(0.95f + 10 * smoothcolor ,0.6f,1.0f);
since the first value in HSB color parameters is used to define the color from the color circle.
Use the smooth coloring algorithm to calculate all of the values within the viewport, then map your palette from the lowest to highest value. Thus, as you zoom in and the higher values are no longer visible, the palette will scale down as well. With the same constants for n and B you will end up with a range of 0.0 to 1.0 for a fully zoomed out set, but at deeper zooms the dynamic range will shrink, to say 0.0 to 0.1 at 200% zoom, 0.0 to 0.0001 at 20000% zoom, etc.
Here is a typical inner loop for a naive Mandelbrot generator. To get a smooth colour you want to pass in the real and complex "lengths" and the iteration you bailed out at. I've included the Mandelbrot code so you can see which vars to use to calculate the colour.
for (ix = 0; ix < panelMain.Width; ix++)
{
cx = cxMin + (double )ix * pixelWidth;
// init this go
zx = 0.0;
zy = 0.0;
zx2 = 0.0;
zy2 = 0.0;
for (i = 0; i < iterationMax && ((zx2 + zy2) < er2); i++)
{
zy = zx * zy * 2.0 + cy;
zx = zx2 - zy2 + cx;
zx2 = zx * zx;
zy2 = zy * zy;
}
if (i == iterationMax)
{
// interior, part of set, black
// set colour to black
g.FillRectangle(sbBlack, ix, iy, 1, 1);
}
else
{
// outside, set colour proportional to time/distance it took to converge
// set colour not black
SolidBrush sbNeato = new SolidBrush(MapColor(i, zx2, zy2));
g.FillRectangle(sbNeato, ix, iy, 1, 1);
}
and MapColor below: (see this link to get the ColorFromHSV function)
private Color MapColor(int i, double r, double c)
{
double di=(double )i;
double zn;
double hue;
zn = Math.Sqrt(r + c);
hue = di + 1.0 - Math.Log(Math.Log(Math.Abs(zn))) / Math.Log(2.0); // 2 is escape radius
hue = 0.95 + 20.0 * hue; // adjust to make it prettier
// the hsv function expects values from 0 to 360
while (hue > 360.0)
hue -= 360.0;
while (hue < 0.0)
hue += 360.0;
return ColorFromHSV(hue, 0.8, 1.0);
}
MapColour is "smoothing" the bailout values from 0 to 1 which then can be used to map a colour without horrible banding. Playing with MapColour and/or the hsv function lets you alter what colours are used.
Seems simple to do by trial and error. Assume you can define HSV1 and HSV2 (hue, saturation, value) of the endpoint colors you wish to use (black and white; blue and yellow; dark red and light green; etc.), and assume you have an algorithm to assign a value P between 0.0 and 1.0 to each of your pixels. Then that pixel's color becomes
(H2 - H1) * P + H1 = HP
(S2 - S1) * P + S1 = SP
(V2 - V1) * P + V1 = VP
With that done, just observe the results and see how you like them. If the algorithm to assign P is continuous, then the gradient should be smooth as well.
My eventual solution was to create a nice looking (and fairly large) palette and store it as a constant array in the source, then interpolate between indexes in it using the smooth coloring algorithm. The palette wraps (and is designed to be continuous), but this doesn't appear to matter much.
What's going on with the color mapping in that image is that it's using a 'log transfer function' on the index (according to documentation). How exactly it's doing it I still haven't figured out yet. The program that produced it uses a palette of 400 colors, so index ranges [0,399), wrapping around if needed. I've managed to get pretty close to matching it's behavior. I use an index range of [0,1) and map it like so:
double value = Math.log(0.021 * (iteration + delta + 60)) + 0.72;
value = value - Math.floor(value);
It's kind of odd that I have to use these special constants in there to get my results to match, since I doubt they do any of that. But whatever works in the end, right?
here you can find a version with javascript
usage :
var rgbcol = [] ;
var rgbcol = MapColor ( Iteration , Zy2,Zx2 ) ;
point ( ctx , iX, iY ,rgbcol[0],rgbcol[1],rgbcol[2] );
function
/*
* The Mandelbrot Set, in HTML5 canvas and javascript.
* https://github.com/cslarsen/mandelbrot-js
*
* Copyright (C) 2012 Christian Stigen Larsen
*/
/*
* Convert hue-saturation-value/luminosity to RGB.
*
* Input ranges:
* H = [0, 360] (integer degrees)
* S = [0.0, 1.0] (float)
* V = [0.0, 1.0] (float)
*/
function hsv_to_rgb(h, s, v)
{
if ( v > 1.0 ) v = 1.0;
var hp = h/60.0;
var c = v * s;
var x = c*(1 - Math.abs((hp % 2) - 1));
var rgb = [0,0,0];
if ( 0<=hp && hp<1 ) rgb = [c, x, 0];
if ( 1<=hp && hp<2 ) rgb = [x, c, 0];
if ( 2<=hp && hp<3 ) rgb = [0, c, x];
if ( 3<=hp && hp<4 ) rgb = [0, x, c];
if ( 4<=hp && hp<5 ) rgb = [x, 0, c];
if ( 5<=hp && hp<6 ) rgb = [c, 0, x];
var m = v - c;
rgb[0] += m;
rgb[1] += m;
rgb[2] += m;
rgb[0] *= 255;
rgb[1] *= 255;
rgb[2] *= 255;
rgb[0] = parseInt ( rgb[0] );
rgb[1] = parseInt ( rgb[1] );
rgb[2] = parseInt ( rgb[2] );
return rgb;
}
// http://stackoverflow.com/questions/369438/smooth-spectrum-for-mandelbrot-set-rendering
// alex russel : http://stackoverflow.com/users/2146829/alex-russell
function MapColor(i,r,c)
{
var di= i;
var zn;
var hue;
zn = Math.sqrt(r + c);
hue = di + 1.0 - Math.log(Math.log(Math.abs(zn))) / Math.log(2.0); // 2 is escape radius
hue = 0.95 + 20.0 * hue; // adjust to make it prettier
// the hsv function expects values from 0 to 360
while (hue > 360.0)
hue -= 360.0;
while (hue < 0.0)
hue += 360.0;
return hsv_to_rgb(hue, 0.8, 1.0);
}