I have a spherical rendering, where the spherical coordinates $\phi$ and $\theta$ are represented by the x and y axis of the image (similar to how world maps work):

enter image description here

Now given a point on the image with the pixel coordinates p($\phi$, $\theta$) I want to calculate the average color of all the neighboring pixels on the sphere within a radius (radius could maybe more easily be defined as an angle, since we are thinking of a unit sphere). For this I need to sum up all of the pixels within the circle, however the circle on the sphere will not be a circle on my map. So the question is, how can I get all the pixel coordinates (= my $\phi$ and $\theta$ angles) within the circle?

Edit: Forgot to describe, how the rendering was done. With $\phi = [0, 2\pi]$ we divide $2\pi$ by the horizontal image resolution. $\theta = [0, \pi]$, so we divide $\pi$ by the vertical image resolution. Using a horizontal resolution that is twice the vertical resolution one pixel represents area on the sphere that is $(\frac{2 * \pi}{res_h}\times\frac{2 * \pi}{res_h})$.

  • $\begingroup$ This very much depends on how your sphere's surface has been mapped to a rectangle. There are many ways to do so and each provides different distortion of areas on the sphere. $\endgroup$ Jan 6, 2022 at 14:55
  • $\begingroup$ @EricTowers sorry for the missing information, I will update the question immediately! $\endgroup$ Jan 6, 2022 at 14:56
  • $\begingroup$ You image is roughly twice as wide as it is high, so your $x$-axis and $y$-axis may be longitude and latitude (with $0$ at the centre). If so, it is worth knowing that, on a sphere, the length of a degree of latitude (north-south) does not vary while the length of a degree of longitude (east-west) is proportional to the cosine of the latitude and would may want to rescale your chosen pixels accordingly. $\endgroup$
    – Henry
    Jan 6, 2022 at 15:06
  • $\begingroup$ @Henry Yes I'm aware of that. Really though the map is just a visualization. In my actual application I have a function, where I plug in phi and theta and I get the corresponding color. Now given a certain phi and theta I want to average color within an area. Question is: which ranges of phi and theta do I need to use $\endgroup$ Jan 6, 2022 at 15:13
  • $\begingroup$ If you want an average that is a valid average by area over the sphere, you also need to assign each pixel a weighting factor, because pixels nearer to the poles actually represent smaller parts of the sphere than pixels near the equator. $\endgroup$
    – David K
    Jan 7, 2022 at 2:59

4 Answers 4


I will stick to your notations $\phi$ for longitude, $\theta$ for latitude, although one finds rather often the inverse convention.

Given: a center $M_0$ (defined by spherical coordinates $(\phi_0,\theta_0)$) and a radius $R$ ($0<R<\pi/2$, measured as an arc on the unit sphere).

The set of points $M$ of the sphere which are interior to the spherical disk with center $M_0$ and radius $R$ is given by the following dot product constraint:


which is equivalent, using classical spherical coordinates, to:

$$\begin{pmatrix}\cos(\phi)\cos(\theta)\\ \sin(\phi)\cos(\theta)\\ \sin(\theta)\end{pmatrix} .\begin{pmatrix}\cos(\phi_0)\cos(\theta_0)\\ \sin(\phi_0)\cos(\theta_0)\\ \sin(\theta_0)\end{pmatrix}>\cos(R)\tag{1}$$

This constraint can be written under the form:


which is an implicit equation in $(\phi,\theta)$ depending upon three parameters $(\phi_0,\theta_0,R)$ that can be visualized with this Geogebra animation (play with the sliders !):


The top sliders $f$ and $t$ refer to the coordinates $\phi_0$ and $\theta_0$ resp. of center $M_0$.

Here are two examples:

enter image description here

Fig. 1: An "ordinary" circle centered in (0,0) with radius $\pi/4$ rendered as a kind of ellipse.

enter image description here

Fig. 2: A "limit case" image of a circle belonging to northern hemisphere, tangent to the equator, (almost) passing throughout North Pole, explaining the almost linear segment ranging approximately from $-\pi/2$ to $\pi/2$.

Remark 1: constraint (1) has been given in the same form by @blamocur.

Remark 2: formula (2) could have been obtained directly by using the spherical law of cosines..

Remark 3: Explicit equations of the circle can be found here.

  • $\begingroup$ I have a little improved my presentation done yesterday (axes' ticks in radians, etc. and also Remark 3). Any comment ? $\endgroup$
    – Jean Marie
    Jan 7, 2022 at 11:10
  • $\begingroup$ Sorry, I commented at a wrong place. $\endgroup$
    – Narasimham
    Jan 7, 2022 at 11:44
  • $\begingroup$ Sorry for the late replies, I'm a bit behind before being able to test the answers out. I will definitely check the answers and tag them :) $\endgroup$ Jan 7, 2022 at 12:16

Let $a$ be the axis of the circle passing through its center, and let $\theta_0$ be the angle of the radius (i.e. $\text{radius} = \sin \theta_0$ )

The $a$ axis has coordinates:

$a = ( \sin \theta_a \cos \phi_a, \sin \theta_a \sin \phi_a, \cos \theta_a)$

you have to find two orthogonal vectors that are orthogonal to $a$, and these can be chosen as follows:

$u_1 = ( \cos \theta_a \cos \phi_a, \cos \theta_a \sin \phi_a , - \sin \theta_a )$

$u_2 = (- \sin \phi_a, \cos \phi_a , 0 )$

The rectangular coordinates of points at the circular region boundary are given by (with respect to the basis $[u_1, u_2, a]$, utilizing the respective spherical coordinates) are:

$(x', y', z') = ( \sin \theta_0 \cos \phi, \sin \theta_0 \sin \phi, \cos \theta_0 )$

where $0 \le \phi \le 2 \pi $

now the world rectangular coordinates of the circular region boundary is

$(x, y, z) = [u_1, u_2, a] (x', y', z') = x' u_1 + y' u_2 + z' a$

once you've calculated $(x, y, z)$ , you can calculate the corresponding $\theta, \phi$ in the original world coordinates, as follows

$ \theta = \cos^{-1} (z)$ and $ \phi = \text{atan2} (x, y) $

By changing $\phi$ over $[0, 2 \pi]$ you get a closed curve in your image, that encloses all the pixels you want.

  • $\begingroup$ Thank you this looks good. I guess I only have to divide up the range to get numerical results, but I'll test this out and mark the answer (prolly until tomorrow :) ). Thanks again! $\endgroup$ Jan 6, 2022 at 15:31
  • $\begingroup$ You're welcome. My pleasure. $\endgroup$ Jan 6, 2022 at 15:41

Let's say you want a neighborhood of a point on a unit square given by angles $\phi_0,\theta_0$ and "angle-radius" $\rho$. We'll assume that the neighborhood does not contain the poles, i.e. $\rho \leq \theta_0 \leq \pi$. You neighborhood can then be described by this inequality:

$$ \langle \mathbf v_{\phi_0,\theta_0} \cdot \mathbf v_{\phi,\theta} \rangle \geq \cos \rho $$

where $\mathbf v_{\phi,\theta}$ is the unit vector with spheric coordinates $\phi,\theta$, and $\langle\cdot\rangle$ denotes dot product. In practice you can scan all points in the $[\phi_0-\rho, \phi_0+\rho] \times [\theta_0-\rho, \theta_0+\rho] $ spheric rectangle and use the above inequality to check whether they belong to the circular area.

Alternately, you can use an equivalent inequality: $$ ||\mathbf v_{\phi,\theta}-\mathbf v_{\phi_0,\theta_0}||^2 \leq 4\sin^2 \frac{\rho}{2}$$


Any circle on a sphere can be obtained as the intersection with a plane. Hence in spherical coordinates, the constraint will be


where $(a,b,c)$ correspond to the normal vector and is also the coordinates of the given point. $d$ is the cosine of the aperture angle of the cone generated by the circle, the radius of which is the sine.

With the point given in spherical coordinates, $$\sin\Phi\cos(\theta-\Theta)\sin\phi+\cos\Phi\cos\phi=\sqrt{1-r^2}$$


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .