$ \def \J {\Delta} \def \half {\frac{1}{2}} \def \kwart {\frac{1}{4}} \def \hieruit {\quad \Longrightarrow \quad} \def \slechts {\quad \Longleftrightarrow \quad} \def \norm {\frac{1}{\sigma \sqrt{2\pi}} \; } \def \EN {\quad \mbox{and} \quad} \def \OF {\quad \mbox{or} \quad} \def \wit {\quad \mbox{;} \quad} \def \spekhaken {\iint} \newcommand{\dq}[2]{\displaystyle \frac{\partial #1}{\partial #2}} \newcommand{\oq}[2]{\partial #1 / \partial #2} \newcommand{\qq}[3]{\frac{\partial^2 #1}{{\partial #2}{\partial #3}}} \def \erf {\operatorname{Erf}} $

Bounding Ellipse

When it comes to the preprocessing of domains and contours, Bounding Ellipses are virtually indespensible. They are far more handsome than bounding boxes of whatever kind in the first place. The only quantities to be determined are the midpoint and the second order variances of the body (or its boundary curve) under consideration. A bounding ellipse is then based upon the inverse tensor of inertia / matrix of variances. The equation of an ellipse of inertia is given by: $$ \left[ \begin{array}{cc} (x-\mu_x) & (y-\mu_y) \end{array} \right] \left[ \begin{array}{cc} \sigma_{xx} & \sigma_{xy} \\ \sigma_{xy} & \sigma_{yy} \end{array} \right]^{-1} \left[ \begin{array}{c} (x-\mu_x) \\ (y-\mu_y) \end{array} \right] = 1 \hieruit $$ $$ \frac{\sigma_{yy} (x-\mu_x)^2 - 2 \sigma_{xy} (x-\mu_x)(y-\mu_y) + \sigma_{xx} (y-\mu_y)^2}{\sigma_{xx}\sigma_{yy}-\sigma_{xy}^2} = E $$ Where $\sigma$ are the variances and $\mu$ are the midpoint coordinates. A bounding ellipse is defined as an ellipse of inertia which its right hand side modified: the constant $1$ is replaced by a another constant $E$. Here $E$ should be adapted in such a way that the whole boundary of the area of interest is contained inside the ellipse. Thus $E$ is the maximum of: $$ \frac{\sigma_{yy} (x_k-\mu_x)^2 - 2 \sigma_{xy} (x_k-\mu_x)(y_k-\mu_y) + \sigma_{xx} (y_k-\mu_y)^2}{\sigma_{xx}\sigma_{yy}-\sigma_{xy}^2} $$ where ($k$) runs through all vertices of the boundary contours.

Best Fit Ellipse

Quite another problem is whether and when an ellipse of inertia can be considered as a Best Fit Ellipse. The latter shall be defined with help of a Least Squares minimization principle, as follows: $$ \sum_k w_k \left[ \frac{\sigma_{yy} (x_k-\mu_x)^2 - 2 \sigma_{xy} (x_k-\mu_x)(y_k-\mu_y) + \sigma_{xx} (y_k-\mu_y)^2}{\sigma_{xx}\sigma_{yy}-\sigma_{xy}^2} - E \right]^2 $$ $= \operatorname{minimum}(E)$ . Differentiating to $E$ simply gives: $$ \sum_k w_k \left[ \frac{\sigma_{yy} (x_k-\mu_x)^2 - 2 \sigma_{xy} (x_k-\mu_x)(y_k-\mu_y) + \sigma_{xx} (y_k-\mu_y)^2}{\sigma_{xx}\sigma_{yy}-\sigma_{xy}^2} - E \right] = 0 $$ If and only if: $$ \sigma_{yy} \sum_k w_k (x_k-\mu_x)^2 - 2 \sigma_{xy} \sum_k w_k (x_k-\mu_x)(y_k-\mu_y) + \sigma_{xx} \sum_k w_k (y_k-\mu_y)^2 $$ $$ = E (\sigma_{xx}\sigma_{yy}-\sigma_{xy}^2) $$ If and only if: $$ \sigma_{yy} \sigma_{xx} - 2 \sigma_{xy} \sigma_{xy} + \sigma_{xx} \sigma_{yy} = 2\,(\sigma_{xx}\sigma_{yy}-\sigma_{xy}^2) = E (\sigma_{xx}\sigma_{yy}-\sigma_{xy}^2) $$ If and only if $E = 2$ . Thus the equation of the Best Fit Ellipse is: $$ \sigma_{yy} (x-\mu_x)^2 - 2 \sigma_{xy} (x-\mu_x)(y-\mu_y) + \sigma_{xx} (y-\mu_y)^2 $$ $$ = 2\,(\sigma_{xx}\sigma_{yy}-\sigma_{xy}^2) $$ Or alternatively: $$ \half\:\frac{\sigma_{yy} (x-\mu_x)^2 - 2 \sigma_{xy} (x-\mu_x)(y-\mu_y) + \sigma_{xx} (y-\mu_y)^2}{\sigma_{xx}\sigma_{yy}-\sigma_{xy}^2} = 1 $$