该方法名为平方差匹配法,计算的公式如式(6.9)所示,这种方法利用平方差来进行匹配,当模板与滑动窗口完全匹配时计算数值为0,两者匹配度越低计算数值越大。
R
(
x
,
y
)
=
∑
x
′
,
y
′
(
T
(
x
′
,
y
′
)
−
I
(
x
+
x
′
,
y
+
y
′
)
)
2
(6.9)
R(x,y) = \sum\limits_{x',y'} {{{(T(x',y') - I(x + x',y + y'))}^2}} \tag{6.9}
R(x,y)=x′,y′∑(T(x′,y′)−I(x+x′,y+y′))2(6.9) 其中 表示模板图像, 表示原图像。
(2)TM_SQDIFF_NORMED:
该方法名为归一化平方差匹配方法,计算公式如式(6.10)所示,这种方法是将平方差方法进行归一化,使得输入结果缩放到了0到1之间,当模板与滑动窗口完全匹配时计算数值为0,两者匹配度越低计算数值越大。
R
(
x
,
y
)
=
∑
x
′
,
y
′
(
T
(
x
′
,
y
′
)
−
I
(
x
+
x
′
,
y
+
y
′
)
)
2
∑
x
′
,
y
′
T
(
x
′
,
y
′
)
2
∗
∑
x
′
,
y
′
I
(
x
+
x
′
,
y
+
y
′
)
2
(6.10)
R(x,y) = \frac{{\sum\limits_{x',y'} {{{(T(x',y') - I(x + x',y + y'))}^2}} }}{{\sqrt {\sum\limits_{x',y'} {T{{(x',y')}^2}} * \sum\limits_{x',y'} {I{{(x + x',y + y')}^2}} } }} \tag{6.10}
R(x,y)=x′,y′∑T(x′,y′)2∗x′,y′∑I(x+x′,y+y′)2x′,y′∑(T(x′,y′)−I(x+x′,y+y′))2(6.10)
(3)TM_CCORR:
该方法名为相关匹配法,计算公式如式(6.11)所示,这类方法采用模板和图像间的乘法操作,所以数值越大表示匹配效果越好,0表示最坏的匹配结果。
R
(
x
,
y
)
=
∑
x
′
,
y
′
(
T
(
x
′
,
y
′
)
∗
I
(
x
+
x
′
,
y
+
y
′
)
)
(6.11)
R(x,y) = \sum\limits_{x',y'} {(T(x',y') * I(x + x',y + y'))} \tag{6.11}
R(x,y)=x′,y′∑(T(x′,y′)∗I(x+x′,y+y′))(6.11)
(4)TM_CCORR_NORMED:
该方法名为归一化相关匹配法,计算公式如式(6.12)所示,这种方法是将相关匹配法进行归一化,使得输入结果缩放到了0到1之间,当模板与滑动窗口完全匹配时计算数值为1,两者完全不匹配时计算结果为0。
R
(
x
,
y
)
=
∑
x
′
,
y
′
(
T
(
x
′
,
y
′
)
∗
I
(
x
+
x
′
,
y
+
y
′
)
)
2
∑
x
′
,
y
′
T
(
x
′
,
y
′
)
2
∗
∑
x
′
,
y
′
I
(
x
+
x
′
,
y
+
y
′
)
2
(6.12)
R(x,y) = \frac{{\sum\limits_{x',y'} {{{(T(x',y') * I(x + x',y + y'))}^2}} }}{{\sqrt {\sum\limits_{x',y'} {T{{(x',y')}^2}} * \sum\limits_{x',y'} {I{{(x + x',y + y')}^2}} } }} \tag{6.12}
R(x,y)=x′,y′∑T(x′,y′)2∗x′,y′∑I(x+x′,y+y′)2x′,y′∑(T(x′,y′)∗I(x+x′,y+y′))2(6.12)
(5)TM_CCOEFF:
该方法名为系数匹配法,计算公式如式(6.13)所示,这种方法采用相关匹配方法对模板减去均值的结果和原图像减去均值的结果进行匹配,这种方法可以很好的解决模板图像和原图像之间由于亮度不同而产生的影响。该方法中模板与滑动窗口匹配度越高计算数值越大,匹配度越低计算数值越小,并且该方法计算结果可以为负数。
R
(
x
,
y
)
=
∑
x
′
,
y
′
(
T
′
(
x
′
,
y
′
)
∗
I
′
(
x
+
x
′
,
y
+
y
′
)
)
(6.13)
R(x,y) = \sum\limits_{x',y'} {(T'(x',y') * I'(x + x',y + y'))} \tag{6.13}
R(x,y)=x′,y′∑(T′(x′,y′)∗I′(x+x′,y+y′))(6.13) 其中:
{
T
′
(
x
′
,
y
′
)
=
T
(
x
′
,
y
′
)
−
1
w
∗
h
∑
x
′
′
,
y
′
′
T
(
x
′
′
,
y
′
′
)
I
′
(
x
+
x
′
,
y
+
y
′
)
=
I
(
x
+
x
′
,
y
+
y
′
)
−
1
w
∗
h
∑
x
′
′
,
y
′
′
I
(
x
+
x
′
′
,
y
+
y
′
′
)
(6.14)
\left\{ {\begin{matrix} {T'(x',y') = T(x',y') - \frac{1}{{w * h}}\sum\limits_{x'',y''} {T(x'',y'')} }\\ {I'(x + x',y + y') = I(x + x',y + y') - \frac{1}{{w * h}}\sum\limits_{x'',y''} {I(x + x'',y + y'')} } \end{matrix}} \right. \tag{6.14}
⎩⎨⎧T′(x′,y′)=T(x′,y′)−w∗h1x′′,y′′∑T(x′′,y′′)I′(x+x′,y+y′)=I(x+x′,y+y′)−w∗h1x′′,y′′∑I(x+x′′,y+y′′)(6.14)
(6)TM_CCOEFF_NORMED:
该方法名为归一化系数匹配法,计算公式如式(6.15)所示,这种方法将系数匹配方法进行归一化,使得输入结果缩放到了1到-1之间,当模板与滑动窗口完全匹配时计算数值为1,当两者完全不匹配时计算结果为-1。
R
(
x
,
y
)
=
∑
x
′
,
y
′
(
T
′
(
x
′
,
y
′
)
∗
I
′
(
x
+
x
′
,
y
+
y
′
)
)
∑
x
′
,
y
′
T
(
x
′
,
y
′
)
2
∗
∑
x
′
,
y
′
I
′
(
x
+
x
′
,
y
+
y
′
)
2
(6.15)
R(x,y) = \frac{{\sum\limits_{x',y'} {(T'(x',y') * I'(x + x',y + y'))} }}{{\sqrt {\sum\limits_{x',y'} {T{{(x',y')}^2} * \sum\limits_{x',y'} {I'{{(x + x',y + y')}^2}} } } }} \tag{6.15}
R(x,y)=x′,y′∑T(x′,y′)2∗x′,y′∑I′(x+x′,y+y′)2x′,y′∑(T′(x′,y′)∗I′(x+x′,y+y′))(6.15)