Statistical Shape Analysis是研究Object的Shape的一门学科。所谓Object的Shape性质是指剔除掉Object因Euclidean Similarity Transformation(移动、旋转和缩放)而变化之后的不变量。如下图所示(第二副图中的三角形是第一副图中三角形向右上移动并顺时旋转 12 0 ∘ 120^{\circ} 120∘的结果,第三幅图中三角形是第一幅图中缩小 1 / 3 1/3 1/3的结果。但它们的Shape是一样的):
Definition 1: Shape is all the geometrical information that remains when location, scale and rotation effects are removed from an object.
An object’s shape is invariant under the Euclidean similarity transformations of translation, scaling and rotation.
现实中研究Object的Shape是通过对Object的采样(Landmarks)进行研究。Landmarks是Object表面一系列的点,点的数量越多,对原Object的“描述”就更加清晰准确。对于Landmarks的选取,一般选择Object面上的角点或者是曲率较高的点。
Definition 2: A landmark is a point of correspondence on each object that matches between and within populations.
There are three basic types of landmarks in our applications: scientific, mathematical and pseudo-landmarks.
A scientific landmark is a point assigned by an expert that corresponds between objects in some scientifically meaningful way, for example the corner of an eye or the meeting of two sutures on a skull. In biological applications such landmarks are also known as anatomical landmarks.
Mathematical landmarks are points located on an object according to some mathematical or geometrical property of the figure, for example at a point of high curvature or at an extreme point.
Pseudo-landmarks are constructed points on an object, located either around the outline or in between scientific or mathematical landmarks. Pseudo-landmarks are useful in matching surfaces, when points can be located on a grid over each surface.
A further type of landmark is the semi-landmark which is a point located on a curve and allowed to slip a small distance in a direction tangent to another corresponding curve.
Object的Size测量对于其Shape研究至关重要,常用的Size测量方法是Centroid Size:
Definition 3: The centroid size is given by:
S ( X ) = ∥ C X ∥ = ∑ i = 1 k ∑ j = 1 m ( X i j − X ˉ j ) 2 , X ∈ ℜ k m , S(X) = \parallel CX \parallel = \sqrt{\sum^{k}_{i=1}\sum^{m}_{j=1}(X_{ij}-\bar{X}_{j})^{2}}, \hspace{0.5cm} X \in \Re^{km}, S(X)=∥CX∥=i=1∑kj=1∑m(Xij−Xˉj)2,X∈ℜkm,
where X i j X_{ij} Xij is the ( i , j ) t h (i,j)th (i,j)th entry of X X X, the arithmetic mean in the jth dimension is X ˉ j = 1 k ∑ i = 1 k X i j \bar{X}_{j} = \frac{1}{k}\sum_{i=1}{k}X_{ij} Xˉj=k1∑i=1kXij,
C = I k − 1 k 1 k 1 k T C = I_{k}-\frac{1}{k}\bf{1}_{k}\bf{1}_{k}^{T} C=Ik−k11k1kT
is the centring matrix,
∥ X ∥ = t r a c e ( X T X ) \parallel X \parallel = \sqrt{trace(X^{T}X)} ∥X∥=trace(XTX)
is the Euclidean norm, I k I_{k} Ik is the k ∗ k k*k k∗k identity matrix (diagonal matrix with ones on the diagonal), and 1 k \bf{1}_{k} 1k is the k ∗ 1 k*1 k∗1 vector of ones.
推导:
X
=
[
c
o
o
r
d
i
n
a
t
e
_
o
f
_
l
a
n
d
m
a
r
k
_
1
c
o
o
r
d
i
n
a
t
e
_
o
f
_
l
a
n
d
m
a
r
k
_
2
⋮
c
o
o
r
d
i
n
a
t
e
_
o
f
_
l
a
n
d
m
a
r
k
_
k
]
k
∗
1
=
[
x
11
…
x
1
m
x
21
…
x
2
m
⋮
⋱
⋮
x
k
1
…
x
k
m
]
k
∗
m
X
ˉ
=
[
1
k
∑
i
=
1
k
x
i
1
1
k
∑
i
=
1
k
x
i
2
…
1
k
∑
i
=
1
k
x
i
m
]
1
∗
m
\begin{aligned} & X = \begin{bmatrix} coordinate\_of\_landmark\_1 \\ coordinate\_of\_landmark\_2 \\ \vdots \\ coordinate\_of\_landmark\_k \\ \end{bmatrix}_{k*1} = \begin{bmatrix} x_{11} & \dots & x_{1m} \\ x_{21} & \dots & x_{2m} \\ \vdots & \ddots & \vdots \\ x_{k1} & \dots & x_{km} \end{bmatrix}_{k*m} \\ \\ & \bar{X} = \begin{bmatrix} \frac{1}{k}\sum_{i=1}^{k}x_{i1} & \frac{1}{k}\sum_{i=1}^{k}x_{i2} & \dots & \frac{1}{k}\sum_{i=1}^{k}x_{im} \end{bmatrix}_{1*m} \end{aligned}
X=⎣⎢⎢⎢⎡coordinate_of_landmark_1coordinate_of_landmark_2⋮coordinate_of_landmark_k⎦⎥⎥⎥⎤k∗1=⎣⎢⎢⎢⎡x11x21⋮xk1……⋱…x1mx2m⋮xkm⎦⎥⎥⎥⎤k∗mXˉ=[k1∑i=1kxi1k1∑i=1kxi2…k1∑i=1kxim]1∗m
X
ˉ
\bar{X}
Xˉ是对
X
X
X在每个Dimension上取均值后所有Landmakrs的“中心点”。如下图所示:
其中,
l
k
l_{k}
lk表示Landmark
k
k
k(
X
[
k
:
]
X[k:]
X[k:])到
X
ˉ
\bar{X}
Xˉ的距离。因此:
∑
i
=
1
k
∑
j
=
1
m
(
X
i
j
−
X
ˉ
j
)
2
=
∑
i
=
1
k
∥
X
i
−
X
ˉ
∥
2
=
∑
i
=
1
k
l
i
2
\sqrt{\sum^{k}_{i=1}\sum^{m}_{j=1}(X_{ij}-\bar{X}_{j})^{2}} = \sqrt{\sum^{k}_{i=1} \parallel X_{i}-\bar{X} \parallel^{2}} = \sqrt{\sum^{k}_{i=1}l_{i}^2}
i=1∑kj=1∑m(Xij−Xˉj)2=i=1∑k∥Xi−Xˉ∥2=i=1∑kli2
即,Centroid Size是类似于由Landmarks组成的“球体”的“半径” 根号下
l
k
l_{k}
lk的平方和。
I
k
=
[
1
0
…
0
0
1
…
0
⋮
⋮
⋱
⋮
0
0
…
1
]
k
∗
k
,
1
k
=
[
1
1
⋮
1
]
C
=
I
k
−
1
k
1
k
1
k
T
=
[
1
−
1
k
0
−
1
k
…
0
−
1
k
0
−
1
k
1
−
1
k
…
0
−
1
k
⋮
⋮
⋱
⋮
0
−
1
k
0
−
1
k
…
1
−
1
k
]
k
∗
k
C
X
=
[
x
11
−
X
ˉ
1
x
12
−
X
ˉ
2
…
x
1
m
−
X
ˉ
m
x
21
−
X
ˉ
1
x
22
−
X
ˉ
2
…
x
2
m
−
X
ˉ
m
⋮
⋮
⋱
⋮
x
k
1
−
X
ˉ
1
x
k
2
−
X
ˉ
2
…
x
k
m
−
X
ˉ
m
]
k
∗
m
\begin{aligned} &I_{k} = \begin{bmatrix} 1 & 0 & \dots & 0 \\ 0 & 1 & \dots & 0 \\ \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \dots & 1 \end{bmatrix}_{k*k}, \bf{1}_{k} = \begin{bmatrix} 1 \\ 1 \\ \vdots \\ 1 \end{bmatrix} \\ \\ &C = I_{k} - \frac{1}{k}\bf{1}_{k}\bf{1}_{k}^{T}= \begin{bmatrix} 1-\frac{1}{k} & 0-\frac{1}{k} & \dots & 0-\frac{1}{k} \\ 0-\frac{1}{k} & 1-\frac{1}{k} & \dots & 0-\frac{1}{k} \\ \vdots & \vdots & \ddots & \vdots \\ 0-\frac{1}{k} & 0-\frac{1}{k} & \dots & 1-\frac{1}{k} \\ \end{bmatrix}_{k*k} \\ \\ &CX = \begin{bmatrix} x_{11}-\bar{X}_{1} & x_{12}-\bar{X}_{2} & \dots & x_{1m}-\bar{X}_{m} \\ x_{21}-\bar{X}_{1} & x_{22}-\bar{X}_{2} & \dots & x_{2m}-\bar{X}_{m} \\ \vdots & \vdots & \ddots & \vdots \\ x_{k1}-\bar{X}_{1} & x_{k2}-\bar{X}_{2} & \dots & x_{km}-\bar{X}_{m} \\ \end{bmatrix}_{k*m} \end{aligned}
Ik=⎣⎢⎢⎢⎡10⋮001⋮0……⋱…00⋮1⎦⎥⎥⎥⎤k∗k,1k=⎣⎢⎢⎢⎡11⋮1⎦⎥⎥⎥⎤C=Ik−k11k1kT=⎣⎢⎢⎢⎡1−k10−k1⋮0−k10−k11−k1⋮0−k1……⋱…0−k10−k1⋮1−k1⎦⎥⎥⎥⎤k∗kCX=⎣⎢⎢⎢⎡x11−Xˉ1x21−Xˉ1⋮xk1−Xˉ1x12−Xˉ2x22−Xˉ2⋮xk2−Xˉ2……⋱…x1m−Xˉmx2m−Xˉm⋮xkm−Xˉm⎦⎥⎥⎥⎤k∗m
C X CX CX的欧几里得范数(向量对应其纯量平方的平方根的映射) ∥ C X ∥ = ∑ i = 1 k ∥ X i − X ˉ ∥ 2 = ∑ i = 1 k ∑ j = 1 m ( X i j − X ˉ j ) 2 \parallel CX \parallel = \sqrt{\sum^{k}_{i=1}\parallel X_{i}-\bar{X} \parallel^{2}} = \sqrt{\sum^{k}_{i=1}\sum^{m}_{j=1}(X_{ij}-\bar{X}_{j})^{2}} ∥CX∥=∑i=1k∥Xi−Xˉ∥2=∑i=1k∑j=1m(Xij−Xˉj)2。