On-Line Geometric Modeling Notes
BÉZIER PATCH SUBDIVISION

Overview

A general method can be specified to subdivide a Bézier patch. This method is specified unlike the matrix methods, as it is based upon the definition of the patch as a set of curves..

To get a pdf version of these notes look here.

The Method for Subdivision

We recall that, if we take the analytic equation of a Bézier patch, fix $u$ and group factors appropriately, we obtain

$${\bf P} (u,v) \: = \: \sum_{j=0}^{m} \left[ \sum_{i=0}^n {\bf P} _{i,j} B_{i,n}(u) \right] B_{j,m}(v)$$

We notice that portion of the equation inside the brackets is the representation of a Bézier curve. If we let ${\bf Q} _j(u)$ be the value inside the brackets, i.e.

$${\bf Q} _j(u) \: = \: \sum_{i=0}^n {\bf P} _{i,j} B_{i,n}(u)$$

Then

$${\bf P} (u,v) \: = \: \sum_{j=0}^{m} {\bf Q} _j(u) B_{j,m}(v)$$

That is, the quantities ${\bf Q} _j(u)$ form the control points of another Bézier curve, and together for all $u$ and $v$, they form the surface.

If, then, we subdivide each of the $m$ rows of the ${\bf P} _{i,j}$ matrix, it implies that the ${\bf Q} _j$s in the above equation represent only points from the first half of the patch (with respect to $u$). The following illustration shows the result of subdividing the rows in the $4 \times 4$ case.

The second half of the patch can be obtained in a similar fashion. The first and second half of the patch, with respect to $v$, can be obtained by subdividing the columns.

Summary

So, using only curve methods, and by subdividing the rows or columns of the control point array, we can effectively subdivide a Bézier patch. This is the most frequently used algorithm in software implementations of subdivision and can be utilized for Bézier patches of arbitrary degree.