Bookstein Shape Coordinates

Shape Coordinates用以描述Object剔除掉Euclidean Similarity Transformation（移动、旋转和缩放）变量后的Shape属性。本文介绍Bookstein Shape Coordinates在二维和三维数据中的应用。

Bookstein-type coordinates for 2D data

Definition 1: Let $(x_j,y_j),j=1,\dots ,k,$ be $k\ge 3$ landmarks in a plane (m = 2 dimensions). Bookstein coordinates $(x_j^B,y_j^B{)}^T,j=3,\dots ,k$, are the remaining coordinates of an object after translating, rotating and rescaling the baseline to $(-\frac{1}{2},0)$ and $(\frac{1}{2},0)$ so that
$$\begin{array}{rl}& x_j^B=[(x_2-x_1)(x_j-x_1)+(y_2-y_1)(y_j-y_1)]/D_{12}^2-\frac{1}{2},\\ & y_j^B=[(x_2-x_1)(y_j-y_1)+(y_2-y_1)(x_j-x_1)]/D_{12}^2,\end{array}$$

where $j=3,\dots ,k$, $D_{12}^2=(x_2-x_1{)}^2+(y_2-y_1{)}^2>0$ and $-\infty <x_j^B,y_j^B<\infty$.

推导：
Step 1: Translating（将$(x_1,y_1{)}^T$移动到$(x_1^B,y_1^B{)}^T=(0,0{)}^T$，得到所有landmarks坐标的平移向量）
$$\left[\begin{array}{c}0\\ 0\end{array}\right]-\left[\begin{array}{c}{x}_1\\ y_1\end{array}\right]=\left[\begin{array}{c}-x_1\\ -y_1\end{array}\right]$$

Step 2: Rotation（借助旋转矩阵Rotation Matrix，将$(x_2,y_2{)}^T$移动到X轴上$(x_2^{\prime },0{)}^T$）
关于原点逆时针旋转$\theta$的矩阵是：
$$\left[\begin{array}{cc}\cos \theta & -\sin \theta \\ \sin \theta & \cos \theta \end{array}\right]$$

那么关于原点顺时针旋转$\theta$的矩阵是：
$$\left[\begin{array}{cc}\cos (-\theta )& -\sin (-\theta )\\ \sin (-\theta )& \cos (-\theta )\end{array}\right]=\left[\begin{array}{cc}\cos \theta & \sin \theta \\ -\sin \theta & \cos \theta \end{array}\right]$$

而$\tan \theta =\frac{y_2-y_1}{x_2-x_1}$。因此，得到Bookstein coordinates的旋转矩阵：
$$\begin{array}{rl}& \left[\begin{array}{cc}\cos \theta & \sin \theta \\ -\sin \theta & \cos \theta \end{array}\right]=\left[\begin{array}{cc}\frac{x_2-x_1}{\sqrt{(x_2-x_1{)}^2+(y_2-y_1{)}^2}}& \frac{y_2-y_1}{\sqrt{(x_2-x_1{)}^2+(y_2-y_1{)}^2}}\\ \frac{y_1-y_2}{\sqrt{(x_2-x_1{)}^2+(y_2-y_1{)}^2}}& \frac{x_2-x_1}{\sqrt{(x_2-x_1{)}^2+(y_2-y_1{)}^2}}\end{array}\right]\\ & =\frac{1}{\sqrt{(x_2-x_1{)}^2+(y_2-y_1{)}^2}}\left[\begin{array}{cc}{x}_2-x_1& y_2-y_1\\ y_1-y_2& x_2-x_1\end{array}\right]\end{array}$$

Step 3: Scaling（将$(x_2^{\prime },0{)}^T$移动到$(x_2^B,y_2^B{)}^T=(1,0{)}^T$，使$(x_1^B,y_1^B{)}^T$到$(x_2^B,y_2^B{)}^T$的距离为1）
$$\frac{1}{\sqrt{(x_2-x_1{)}^2+(y_2-y_1{)}^2}}$$

因此，
$$\begin{array}{rl}& \left[\begin{array}{c}{x}_j^B\\ y_j^B\end{array}\right]=Scaling\circ Rotation\circ Translating(\left[\begin{array}{c}{x}_j\\ y_j\end{array}\right])\\ & =\frac{1}{(x_2-x_1{)}^2+(y_2-y_1{)}^2}\left[\begin{array}{cc}{x}_2-x_1& y_2-y_1\\ y_1-y_2& x_2-x_1\end{array}\right](\left[\begin{array}{c}{x}_j\\ y_j\end{array}\right]+\left[\begin{array}{c}-x_1\\ -y_1\end{array}\right])\\ & =\frac{1}{(x_2-x_1{)}^2+(y_2-y_1{)}^2}\left[\begin{array}{c}(x_2-x_1)(x_j-x_1)+(y_2-y_1)(y_j-y_1)\\ (y_1-y_2)(x_j-x_1)+(x_2-x_1)(y_j-y_1)\end{array}\right]\end{array}$$

Step 4: 为了对称性，可将$(x_j^B,y_j^B{)}^T$左移$\frac{1}{2}$。

---

Bookstein-type coordinates for 3D data

Step 1: Translating（将$(x_1,y_1,z_1{)}^T$移动到$(x_1^B,y_1^B,z_1^B{)}^T=(0,0,0{)}^T$，得到所有landmarks坐标的平移向量）
$$\left[\begin{array}{c}0\\ 0\\ 0\end{array}\right]-\left[\begin{array}{c}{x}_1\\ y_1\\ z_1\end{array}\right]=\left[\begin{array}{c}-x_1\\ -y_1\\ -z_1\end{array}\right]$$

Step 2: Rotation（借助旋转矩阵Rotation Matrix，将$(x_2,y_2,z_2{)}^T$移动到X轴上$(x_2^{\prime },0,0{)}^T$，并将$(x_3,y_3,z_3{)}^T$移动到$X-Y$平面上$(x_3^{\prime },y_3^{\prime },0{)}^T$）

$(x2-x1,y2-y1,z2-z1{)}^T$绕$X$轴（在$Y-Z$平面）顺时针旋转：
$$R_x({\theta }_x)=\left[\begin{array}{ccc}1& 0& 0\\ 0& \cos {\theta }_x& \sin {\theta }_x\\ 0& -\sin {\theta }_x& \cos {\theta }_x\end{array}\right]$$

$(x2-x1,y2-y1,z2-z1{)}^T$绕$Y$轴（在$X-Z$平面）顺时针旋转：
$$R_y({\theta }_y)=\left[\begin{array}{ccc}\cos {\theta }_y& 0& -\sin {\theta }_y\\ 0& 1& 0\\ \sin {\theta }_y& 0& \cos {\theta }_y\end{array}\right]$$

$(x2-x1,y2-y1,z2-z1{)}^T$绕$Z$轴（在$X-Y$平面）顺时针旋转：
$$R_z({\theta }_z)=\left[\begin{array}{ccc}\cos {\theta }_z& \sin {\theta }_z& 0\\ -\sin {\theta }_z& \cos {\theta }_z& 0\\ 0& 0& 1\end{array}\right]$$

Step 3: Scaling（将$(x_2^{\prime },0,0{)}^T$移动到$(x_2^B,y_2^B,z_2^B{)}^T=(1,0,0{)}^T$，使$(x_1^B,y_1^B,z_1^B{)}^T$到$(x_2^B,y_2^B,z_2^B{)}^T$的距离为1）

$$\frac{1}{\sqrt{(x_2-x_1{)}^2+(y_2-y_1{)}^2+(z_2-z_1{)}^2}}$$

因此，
$$\left[\begin{array}{c}{x}_j^B\\ y_j^B\\ z_j^B\end{array}\right]=Scaling\circ R_x({\theta }_x)\circ R_y({\theta }_y)\circ R_z({\theta }_z)\circ Translating(\left[\begin{array}{c}{x}_j\\ y_j\\ z_j\end{array}\right])$$