One possible parameterisation of the sphere will be discussed along with the transformation required to texture map a sphere. An angle parameterisation of the sphere is

x = r sin(theta) cos(phi)
y = r sin(theta) sin(phi)
z = r cos(theta)

where r is the radius, theta the angle from the z axis (0 <= theta <= pi), and phi the angle from the x axis (0 <= phi <= 2pi). Textures are conventionally specified as rectangular images which are most easily parameterised by two cartesian type coordinates (u,v) say, where 0 <= u,v <= 1. The equation above for the sphere can be rewritten in terms of u and v as

x = r sin(v pi) cos(u 2 pi)
y = r sin(v pi) sin(u 2 pi)
z = r cos(v pi)

Solving for the u and v from the above gives

v = arccos(z/r) / pi
u = ( arccos(x/(r sin(v pi))) ) / (2 pi)

So, given a point (x,y,z) on the surface of the sphere the above gives the point (u,v) each component of which can be appropriately scaled to index into a texture image.

Note

• A sphere cannot be "unwrapped" without distortion, for example, the length between points on the sphere will not equal the distance between points on the unwrapped plane.

• When implementing this in code it is important to note that most implementations of arccos() returns value from 0 to pi and not 0 to 2 pi as the the formula above assumes. The second half cycle of the arccos function is obtained by noticing the sign of the y value. So the transformation written in C might be as follows

#define PI 3.141592654
#define TWOPI 6.283185308

void SphereMap(x,y,z,radius,u,v)
double x,y,z,r,*u,*v;
{
   *v = acos(z/radius) / PI;
   if (y >= 0)
      *u = acos(x/(radius * sin(PI*(*v)))) / TWOPI;
   else
      *u = (PI + acos(x/(radius * sin(PI*(*v))))) / TWOPI;
}

There are still two special points, the exact north and south poles of the sphere, each of these two points needs to be "spread" out along the whole edge v=0 and v=1. In the formula above this is where sin(v pi) = 0.

While straightforward many people seem to have trouble creating a sphere with texture coordinates. Here's the way I do it (written for clarity rather than efficiency).

Note

• The whole line at the North pole and the South pole texture map onto a single point at the poles.

• While this linear mapping of lines of latitude is fine for general textures, it may not be correct for particular image textures such as maps of the Earth. In those cases the latitude texture coordinates need to be matched to the latitude function used to make the image.
  

  

• On many implementations triangle strips are muck efficient than quad strips. On the other hand triangle strips don't look so good in wireframe mode. Depending on personal taste the line glBegin(GL_QUAD_STRIP); can be replaced with glBegin(GL_TRIANGLE_STRIP);

/*
   Create a sphere centered at c, with radius r, and precision n
   Draw a point for zero radius spheres
*/
void CreateSphere(XYZ c,double r,int n)
{
   int i,j;
   double theta1,theta2,theta3;
   XYZ e,p;

   if (r < 0)
      r = -r;
   if (n < 0)
      n = -n;
   if (n < 4 || r <= 0) {
      glBegin(GL_POINTS);
      glVertex3f(c.x,c.y,c.z);
      glEnd();
      return;
   }

   for (j=0;j<n/2;j++) {
      theta1 = j * TWOPI / n - PID2;
      theta2 = (j + 1) * TWOPI / n - PID2;

      glBegin(GL_QUAD_STRIP);
      for (i=0;i<=n;i++) {
         theta3 = i * TWOPI / n;

         e.x = cos(theta2) * cos(theta3);
         e.y = sin(theta2);
         e.z = cos(theta2) * sin(theta3);
         p.x = c.x + r * e.x;
         p.y = c.y + r * e.y;
         p.z = c.z + r * e.z;

         glNormal3f(e.x,e.y,e.z);
         glTexCoord2f(i/(double)n,2*(j+1)/(double)n);
         glVertex3f(p.x,p.y,p.z);

         e.x = cos(theta1) * cos(theta3);
         e.y = sin(theta1);
         e.z = cos(theta1) * sin(theta3);
         p.x = c.x + r * e.x;
         p.y = c.y + r * e.y;
         p.z = c.z + r * e.z;

         glNormal3f(e.x,e.y,e.z);
         glTexCoord2f(i/(double)n,2*j/(double)n);
         glVertex3f(p.x,p.y,p.z);
      }
      glEnd();
   }
}

It is a ssmall modification to enable one to create subsets of a sphere....3 dimensional wedges. As an example see the following code.

/*
   Create a sphere centered at c, with radius r, and precision n
   Draw a point for zero radius spheres
   Use CCW facet ordering
   "method" is 0 for quads, 1 for triangles
      (quads look nicer in wireframe mode)
   Partial spheres can be created using theta1->theta2, phi1->phi2
   in radians 0 < theta < 2pi, -pi/2 < phi < pi/2
*/
void CreateSphere(XYZ c,double r,int n,int method,
   double theta1,double theta2,double phi1,double phi2)
{
   int i,j;
   double t1,t2,t3;
   XYZ e,p;

   /* Handle special cases */
   if (r < 0)
      r = -r;
   if (n < 0)
      n = -n;
   if (n < 4 || r <= 0) {
      glBegin(GL_POINTS);
      glVertex3f(c.x,c.y,c.z);
      glEnd();
      return;
   }

   for (j=0;j<n/2;j++) {
      t1 = phi1 + j * (phi2 - phi1) / (n/2);
      t2 = phi1 + (j + 1) * (phi2 - phi1) / (n/2);

      if (method == 0)
         glBegin(GL_QUAD_STRIP);
      else
         glBegin(GL_TRIANGLE_STRIP);

      for (i=0;i<=n;i++) {
         t3 = theta1 + i * (theta2 - theta1) / n;

         e.x = cos(t1) * cos(t3);
         e.y = sin(t1);
         e.z = cos(t1) * sin(t3);
         p.x = c.x + r * e.x;
         p.y = c.y + r * e.y;
         p.z = c.z + r * e.z;
         glNormal3f(e.x,e.y,e.z);
         glTexCoord2f(i/(double)n,2*j/(double)n);
         glVertex3f(p.x,p.y,p.z);

         e.x = cos(t2) * cos(t3);
         e.y = sin(t2);
         e.z = cos(t2) * sin(t3);
         p.x = c.x + r * e.x;
         p.y = c.y + r * e.y;
         p.z = c.z + r * e.z;
         glNormal3f(e.x,e.y,e.z);
         glTexCoord2f(i/(double)n,2*(j+1)/(double)n);
         glVertex3f(p.x,p.y,p.z);

      }
      glEnd();
   }
}

When texture mapping a sphere with a rectangular texture image with polar texture coordinates, the parts of the image near the poles get distorted. Given the different topology between a sphere and plane, there will always be some nonlinear distortion or cut involved. This normally manifests itself in pinching at the poles where rows of pixels are being compressed tighter and tighter together the closer one gets to the pole. At the poles is the extreme case where the whole top and bottom row of pixels in the texture map is compressed down to one point. The spherical images below on the left are examples of this pinching using the rectangular texture also on the left.

It is simple to correct for this by distorting the texture map. Assume the texture map is mapped vertically onto lines of latitude (theta) and mapped horizontally onto lines of longitude (phi). There is no need to modify theta but phi is scaled as we approach the two poles by cos(theta). The diagram below illustrates the conventions used here.

The pseudo-code for this distortion might be something like the following, note that the details of how to create and read the images are left up to your personal preferences.

   double theta,phi,phi2;
   int i,i2,j;
   BITMAP *imagein,*imageout;

   Form the input and output image arrays
   Read an input image from a file

   for (j=0;j<image.height;j++) {
      theta = PI * (j - (image.height-1)/2.0) / (double)(image.height-1);
      for (i=0;i<image.width;i++) {
         phi  = TWOPI * (i - image.width/2.0) / (double)image.width;
         phi2 = phi * cos(theta);
         i2  = phi2 * image.width / TWOPI + image.width/2;
         if (i2 < 0 || i2 > image.width-1) {
            newpixel = red;                         /* Should not happen */
         } else {
            newpixel = imagein[j*image.width+i2];
         }
         imageout[j*image.width+i] = image.newpixel;
      }
   }

   Do something with the output image

Applying this transformation to a regular grid is show below.

Perhaps a more illustrative example is given below for a "moon" texture. Note that in general if the texture tiles vertically and horizontally then after this distortion it will no longer tile horizontally. A number of tiling methods can be used to correct for this. One is to replicate and mirror the texture horizontally, since the distortion is symmetric about the horizontal center line of the image, the result will tile horizontally. Another method is to overlap two copies of the texture after the distortion with appropriate masks that fade the appropriate texture out at the nontiling borders.

This final example illustrates how after the distortion the image detail is evenly spread over the spherical object instead of acting like lines of longitude that get closer near the poles.

The following lists some of the most common texture mapping methods use by raytracing/rendering engines. The methods are illustrated by mapping the simple rectangular tile texture on the right onto a cube, sphere, and cylinder. Repeated tiling is used in all cases and where possible the tiling scale factors are kept constant.

Planar

Cubic

Cylindrical

Rectangular Cylindrical

Spherical

Miscellaneous examples

Logo for the The Australasian Society for Psychophysiology, Inc.

VRML version of the cover for Neuroimage 8. For a paper entitled "Steady State Visually Evoked Potential Correlates of Auditory Hallucinations in Schizophrenia".