数字图像处理MATLAB学习笔记（四）

数字图像处理MATLAB学习笔记（四）

Geometric Transformations and Image Processing

1 Transforming Points

点变换就是进行空间坐标系统的转换，即将一个在空间a的点映射到空间b中去。

我们可以将这种输入到输出的正向的空间映射用公式表达为： ( x , y ) = T { ( w , z } {(x,y)=T\{(w,z\}} (x,y)=T{(w,z}。如果可以反向映射的话，表达式满足 ( u , z ) = T − 1 { ( x , y ) } {(u,z)=T^{-1}\{(x,y)\}} (u,z)=T−1{(x,y)}。

图像的geometric transformations(几何变换)也就被定为geometric coordinate transformations(几何坐标变换)。我们令$f(w,z)$表示输入空间的一幅图像，$g(x,y)$表示输出空间的一幅图像，那么根据上述映射关系，我么可以得到： g ( x , y ) = f ( T − 1 { ( x , y ) } ) {g(x,y)=f(T^{-1}\{(x,y)\})} g(x,y)=f(T−1{(x,y)})。

在MATLAB中，我们使用function maketform创建转换形式变量tform：

tform = maketform(transform_type, params, ...)

按照顺序来，先说一下custom：

tform = maketform('custom', ndims_in, ndims_out, forward_fcn, inv_function, tdata)

其中，tdata是关于forward_fcn和inv_function的系数。

Example Creating a custom tform structure and using it to transform points：

这里我们先设定一下输入与输出空间之间的映射关系方程为：
( x , y ) = T { ( w , z ) } = ( 3 w , 2 z ) ( w , z ) = T − 1 { ( x , y ) } = ( x / 3 , y / 2 ) (x,y)=T\{(w,z)\}=(3w,2z)\\ (w,z)=T^{-1}\{(x,y)\}=(x/3,y/2) (x,y)=T{(w,z)}=(3w,2z)(w,z)=T−1{(x,y)}=(x/3,y/2)
那么，我们可以根据上式编写代码如下：

>> forward_fcn = @(wz, tdata) [3*wz(:,1), 2*wz(:,2)];
>> inverse_fcn = @(xy, tdata) [xy(:,1)/3, xy(:,2)/2];
>> tform1 = maketform('custom', 2, 2, forward_fcn, inverse_fcn, [])
tform1 = 
  struct with fields:
       ndims_in: 2
      ndims_out: 2
    forward_fcn: @(wz,tdata)[3*wz(:,1),2*wz(:,2)]
    inverse_fcn: @(xy,tdata)[xy(:,1)/3,xy(:,2)/2]
          tdata: []

这里我们做一下简单的测试，并使用工具箱中的function tformfwd和function tforminv进行正向和反向映射：

WZ = [1 1; 3 2];
>> XY = tformfwd(WZ, tform1)
XY =
     3     2
     9     4
>> WZ2 = tforminv(XY, tform1)
WZ2 =
     1     1
     3     2

我们再设定一个输入与输出空间之间的映射关系：
( x , y ) = T { ( w , z ) } = ( w + 0.4 z , z ) ( w , z ) = T − 1 { ( x , y ) } = ( x − 0.4 y , y ) (x,y)=T\{(w,z)\}=(w+0.4z,z)\\ (w,z)=T^{-1}\{(x,y)\}=(x-0.4y,y) (x,y)=T{(w,z)}=(w+0.4z,z)(w,z)=T−1{(x,y)}=(x−0.4y,y)
我们比着葫芦画瓢，代码很容易实现：

>> forward_fcn2 = @(wz, tdata) [wz(:,1)+wz(:,2)*0.4, wz(:,2)];
>> inverse_fcn2 = @(xy, tdata) [xy(:,1)-xy(:,2)*0.4, xy(:,2)];
>> tform2 = maketform('custom', 2, 2, forward_fcn2, inverse_fcn2, []);
>> WZ
WZ =
     1     1
     3     2
>> XY = tformfwd(WZ, tform2)
XY =
    1.4000    1.0000
    3.8000    2.0000
>> WZ2 = tforminv(XY, tform2)
WZ2 =
    1.0000    1.0000
    3.0000    2.0000

我们根据上述概念，自定义两个M-function pointgrid和vistform，以便于目测并检验给定的变换

function pointgrid：

function wz = pointgrid(corners)
%POINTGRID Points arranged on a grid.
%   WZ = POINTGRI D (CORNERS) computes a set point of points on a
%   grid containing 10 horizontal and vertical lines. Each line
%   contains 50 points. CORNERS is a 2-by-2 matrix. The first
%   row contains the horizontal and vertical coordinates of one
%   corner of the grid. The second row contains the coordinates
%   of the opposite corner. Each row of the P-by-2 output
%   matrix, WZ, contains the coordinates of a point on the output
%   grid.

% Create 10 horizontal lines containing 50 points each.
[w1, z1] = meshgrid(linspace(corners(1,1), corners(2,1), 46), ...
    linspace(corners(1), corners(2), 10));

% Create 10 vertical lines containing 50 points each.
[w2, z2] = meshgrid(linspace(corners(1,1), corners(2,1), 10), ...
    linspace(corners(1), corners(2), 46));

% Create a P-by-2 matrix containing all the input-space points.
wz = [w1(:), z1(:); w2(:), z2(:)];

function vistform：

function vistform(tform, wz)
%VISTFORM Visualization transformation effect on set of points.
%   VISTFORM(TFORM, WZ) shows two plots. On the left are the
%   points in each row of the P-by-2 matrix WZ. On the right are
%   the spatially transformed points using TFORM.

% Transform the points to output space.
xy = tformfwd(tform, wz);

% Compute axes limits for both plots. Bump the limites outward slightly.
minlim = min([wz; xy], [], 1);
maxlim = max([wz; xy], [], 1);
bump = max((maxlim - minlim) * 0.05, 0.1);
limits = [minlim(1)-bump(1), maxlim(1)+bump(1), ...
    minlim(2)-bump(2), maxlim(2)+bump(2)];

subplot(1,2,1)
grid_plot(wz, limits, 'w', 'z');

subplot(1,2,2)
grid_plot(xy, limits, 'x', 'y');

%------------------------------------------------------------
function grid_plot(ab, limits, a_label, b_label)
plot(ab(:,1), ab(:,2), '.', 'MarkerSize', 2)
axis equal, axis ij, axis(limits);
set(gca, 'XAxisLocation', 'top')
xlabel(a_label), ylabel(b_label)

我们测试一下：

>> vistform(tform1, pointgrid([0 0; 100 100]))
>> figure, vistform(tform2, pointgrid([0 0; 100 100]))

2 Affine Transformations仿射变换

这里我们把之前提到的映射方程给用矩阵进行表示：
$$\begin{gathered} [x,y]=[w,z]\left[\begin{array}{cc}{a}_{11}& a_{12}\\ a_{21}& a_{22}\end{array}\right]+[b_1,b_2] \\ 为了方便处理b，所以空间坐标矩阵后都加上一个元素，值为1，化简为： \\ [x,y,1]=[w,z,1]\left[\begin{array}{ccc}{a}_{11}& a_{12}& 0\\ a_{21}& a_{22}& 0\\ b_1& b_2& 0\end{array}\right] \\ 我们就可以令T=\left[\begin{array}{ccc}{a}_{11}& a_{12}& 0\\ a_{21}& a_{22}& 0\\ b_1& b_2& 0\end{array}\right]，则上述公式简化为[x,y,1]=[w,z,1]T \end{gathered}$$
其中，$T$矩阵也称为affine matrix仿射矩阵，这里也就是affine参数起作用了。

Example：

>> T = [1 0 0; 0.4 1 0; 0 0 1];
>> tform3 = maketform('affine', T);
>> WZ = [1 1; 3 2];
>> XY = tformfwd(WZ, tform3)
XY =
    1.4000    1.0000
    3.8000    2.0000

那么，affine matrix的一个特殊形式如下：
$$\begin{gathered} T=\left[\begin{array}{ccc}s\cos \theta & s\sin \theta & 0\\ -s\sin \theta & s\cos \theta & 0\\ b_1& b_2& 1\end{array}\right] \\ OR \\ T=\left[\begin{array}{ccc}s\cos \theta & s\sin \theta & 0\\ s\sin \theta & -s\cos \theta & 0\\ b_1& b_2& 1\end{array}\right] \end{gathered}$$
这些，类似于旋转、平移和反射的，都属于affine transformations的子集，被称为similarity transformations。

我们简单罗列一下affine matrix对应的transformations：

Type	affine matrix	coordinate equations
Identity	$\left[\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& 1\end{array}\right]$	$$\begin{gathered} x=w \\ y=z \end{gathered}$$
Scaling	$\left[\begin{array}{ccc}{s}_x& 0& 0\\ 0& s_y& 0\\ 0& 0& 1\end{array}\right]$	$$\begin{gathered} x=s_{x}w \\ y=s_{y}z \end{gathered}$$
Rotation	$\left[\begin{array}{ccc}s\cos \theta & s\sin \theta & 0\\ -s\sin \theta & s\cos \theta & 0\\ 0& 0& 1\end{array}\right]$	$$\begin{gathered} x=w\cos \theta -z\sin \theta \\ y=w\sin \theta +z\cos \theta \end{gathered}$$
Shear(horizontal)	$\left[\begin{array}{ccc}1& 0& 0\\ \alpha & 1& 0\\ 0& 0& 1\end{array}\right]$	$$\begin{gathered} x=w+\alpha z \\ y=z \end{gathered}$$
Shear(vertical)	$\left[\begin{array}{ccc}1& 0& 0\\ \beta & 1& 0\\ 0& 0& 1\end{array}\right]$	$$\begin{gathered} x=w \\ y=\beta w+z \end{gathered}$$
Vertical reflection	$\left[\begin{array}{ccc}1& 0& 0\\ 0& -1& 0\\ 0& 0& 1\end{array}\right]$	$$\begin{gathered} x=w \\ y=-z \end{gathered}$$
Translation	$\left[\begin{array}{ccc}1& 0& 0\\ 0& -1& 0\\ {\delta }_x& {\delta }_y& 1\end{array}\right]$	$$\begin{gathered} x=w+{\delta }_x \\ y=z+{\delta }_y \end{gathered}$$

3 Projection Transformations

投影变换属于几何变换的一种，这里与affine transfromations不同的是，辅助坐标不是constant。表达式如下：
$$\begin{gathered} [x^{\prime },y^{\prime },h]=[w,z,1]\left[\begin{array}{ccc}{a}_{11}& a_{12}& a_{13}\\ a_{21}& a_{22}& a_{23}\\ b_1& b_2& 1\end{array}\right] \\ Conditions:a_{13}\ne 0,a_{23}\ne 0,x=x^{\prime }/h,y=y^{\prime }/h \end{gathered}$$
使用的参数时projective。Example：

>> T = [-2.7391 0.2928 -0.6376
         0.7426 -0.7501 0.8088
         2.8750 0.7500 1.0000];
>> tform = maketform('projective', T);
>> vistform(tform, pointgrid([0 0; 1 1]));

4 Applying Geometric Transformations to Images

逆映射使用function imtransform，语法如下：

g = imtransform(f, tform)

下面进行一些应用练习：

Example Geometric transformations of images：

affine transform遵循的scaling affine matrix为$T=\left[\begin{array}{ccc}{s}_x& 0& 0\\ 0& s_y& 0\\ 0& 0& 1\end{array}\right]$，rotation affine matrix为$T=\left[\begin{array}{ccc}s\cos \theta & s\sin \theta & 0\\ -s\sin \theta & s\cos \theta & 0\\ 0& 0& 1\end{array}\right]$

>> f = checkerboard(50);
>> sx = 0.75;
>> sy = 1.25;
>> T = [sx 0 0; 0 sy 0; 0 0 1];
>> t1 = maketform('affine', T);
>> g1 = imtransform(f, t1);
>> theta = pi/6;
>> T2 = [cos(theta) sin(theta) 0; -sin(theta) cos(theta) 0; 0 0 1];
>> t2 = maketform('affine', T2);
>> g2 = imtransform(f, t2);
>> T3 = [0.5768 0.0782 -0.0009; 0.0127 0.5371 -0.0009; 0.5987 0.5761 1.0000];
>> tform3 = maketform('projective', T3);
>> g3 = imtransform(f, tform3);
>> subplot(221);imshow(f);title('original image');xlabel('(a)');
>> subplot(222);imshow(g1);title('scaling image');xlabel('(b)');
>> subplot(223);imshow(g2);title('rotation image');xlabel('(c)');
>> subplot(224);imshow(g3);title('projective image');xlabel('(d)');

5 Image Coordinate Systems in MATLAB

MATLAB的坐标系统：这里强调一下，MATLAB中的image coordinate的原点在左上角，为了方便查看像素点位置坐标，我们可以使用如下语句，使得下回每一次使用imshow显示图像时，图像上都可以显示坐标轴

>> iptsetpref imshowAxesVisible on

5.1 Output Image Location

我们可以通过imtransform获得图片反映射后的坐标信息：

[g, xdata, ydata] = imtransform(f, tform, 'FillValue', 255)

Note：imtransform中的FillerValue指的是对于任何输出图像像素的值，对应输入图像边界外的输入空间位置，默认情况下为0。

Example Displaying input and output images together in a common coordinate system：

>> theta = 3*pi/4;
>> T = [cos(theta) sin(theta) 0; -sin(theta) cos(theta) 0; 0 0 1];
>> tform = maketform('affine', T);
>> [g, xdata, ydata] = imtransform(f, tform, 'FillValue', 255);
>> imshow(f)
>> hold on
>> imshow(g, 'XData', xdata, 'YData', ydata)
>> axis auto
>> axis on
>> T = [1 0 0; 0 1 0; 1300 200 1];
>> tform = maketform('affine', T);
>> [g, xdata, ydata] = imtransform(f, tform, 'FillValue', 255);
>> imshow(f)
>> hold on
>> imshow(g, 'XData', xdata, 'YData', ydata)
>> axis on
>> axis auto

5.2 Controlling the Output Grid

imtransform在输出空间的定位与计算步骤：

1. 决定输入image得边界矩形
2. 在边界矩形上把点映射到输出空间
3. 计算变换后输出空间中点的边界矩形
4. 计算位于输出空间中、边界矩形内的网格上的输出图像像素

这里说一下imtransform2函数，实际上是对imtransform函数的一个封装，可以直接对输入与输出的图像进行比较。

function g = imtransform2(f, varargin)
%IMTRANSFORM2 2-D image transformation with fixed output location
%   G = IMTRANSFORM2(F, TFORM, ...) applies a 2-D geometric
%   transformation to an image. IMTRANSFORM2 fixes the output image
%   location to cover the same region as the input image.
%   IMTRANSFORM2 takes the same set of optional parameter/value 
%   pairs as IMTRANSFORM.

[M, N] = size(f);
xdata = [1 N];
ydata = [1 M];
g = imtransform(f, varargin{:}, 'XData', xdata, 'YData', ydata);

6 Image Interpolation 图像内插

内插：用离散数据构建连续定义的函数

通常有两个步骤：

1. 连续数据离散化，即连续域到离散域的映射，通常要用到相应的内插核函数。常见的内插核有：
   1. 盒状核：$h_B(x)=\{\begin{array}{ll}1& -0.5\le x\le 0.5\\ 0& \text{otherwise}\end{array}$
   2. 三角核：$h_T(x)=\{\begin{array}{ll}1-|x|& x\le 1\\ 0& \text{otherwise}\end{array}$
   3. 立方核：$h_C(x)=\{\begin{array}{ll}1.5{|x|}^3-2.5{|x|}^2+1& |x|\le 1\\ -0.5{|x|}^3+2.5{|x|}^2-4|x|+2& 1<|x|\le 2\\ 0& \text{otherwise}\end{array}$
2. 在离散位置估计函数值。

6.1 Interpolation in Two Dimensions

常用的二维内插方法是把二维内插问题化为多个一维内插任务。

例如：求$f(x,y)$先行内插的$f^{\prime }(2.6,1.4)$

1. 用$f(2,1)$和$f(3,1)$之间的线性内插决定$f^{\prime }(2.6,1)$
2. 用$f(2,2)$和$f(3,2)$之间的线性内插决定$f^{\prime }(2.6,2)$
3. 用$f^{\prime }(2.6,1)$和$f^{\prime }(2.6,2)$之间的线性内插决定$f^{\prime }(2.6,1.4)$

6.2 Comparing Interpolation Methods

我们自定义一个function reprotate，这个函数的作用是比较计算速度和图像质量（我们用timeit记录时间）

我们通过使用reprotate比较nearest、bilinear、bicubic三种插值方法的效果，图片进行十二次旋转输入。

function g = reprotate(f, interp_method)
%REPROTATE Rotate image repeatedly
%    G = REPROTATE (F, INTERP METHOD) rotates the input image, F,
%    twelve times in succession as a test of di fferent inte rpolation
%    methods. INTERP METHOD can be one of the strings nearest' ,
%    'bilinear' , or 'bicubic'.

% Form a spatial transformation that rotates the image about its
% center point. The trans formation is formed as a composite of
% three affine trans formations :
% 
% 1. Transform the center of the image to the origin.
center = fliplr(1 + size(f) / 2);
A1 = [1 0 0; 0 1 0; -center, 1];

% 2. Rotate 30 degrees about the origin
theta = 30*pi/180;
A2 = [cos(theta) -sin(theta) 0; sin(theta) cos(theta) 0; 0 0 1];

% 3. Transform from the origin back to the original center location.
A3 = [1 0 0; 0 1 0; center, 1];
% Compose the three transforms using matrix multiplication
A = A1 * A2 * A3;
tform = maketform('affine', A);

% Apply the rotation 12 times in sequence. Use imtransform2 so that
% each successive transformation is computed using the same location
% and size as the original image.
g = f;
for k = 1:12
    g = imtransform2(g, tform, interp_method);
end

Example Comparing speed and image quality for several interpolation methods：

>> f = imread('Fig0517(a).tif');
>> timeit(@() reprotate(f, 'nearest'))
ans =
    0.0585
>> timeit(@() reprotate(f, 'bilinear'))
ans =
    0.0918
>> timeit(@() reprotate(f, 'bicubic'))
ans =
    0.1676

我们可以发现，nearest最近邻内插值速度最快，其次是bilinear双线性插值，bicubic双三次内插最慢。效果图对比：

最终，我们可以发现，nearest呈锯齿状，bilinear的效果最差，图像中心较为模糊，bicubic虽然运行时间最慢，但是效果最好，具有更为平滑的边缘和更少的模糊。

7 Image Registration图像配准

7.1 The Registration Process

图像配准处理的基本操作主要如下：

1. 检测特征
2. 匹配相应的特征
3. 推断/确定几何变换
4. 通过几何变换用另一幅图像对准一幅图像

7.2 Manual Feature Selection and Matching Using cpselect

使用function cpselect手动筛选特征点进行匹配，使用规则如下：

>> cpselect(FIGURE1, FIGURE2)

7.3 Inferring Transformation Parameters Using cp2tform 使用cp2tform推断变换参数

cp2tform的语法为：

tform = cp2tform(input_points, base_points, transformtype)

其中，input_points和base_points是两个Px2大小的包含相应特征的矩阵。transformtype是所希望的变换类型。对应的tform类型和可以生成的maketform或cp2tform。其中，maketform用于直接指定变换参数，cp2tform用于相应的特征位置对估计变换参数。

Type of Transformation	Description	Functions
assine	缩放、旋转、剪切和平移的结合；直线保持为直线，平行线保持平行	maketform、cp2tform
box	独立地沿着每一维缩放和平移，仿射的子集	maketform
composite复合	几何变换的集合，可顺序地使用	maketform
custom定制	用户定义的几何变换，函数自定义	maketform
LWM	局部加权平均，局部变化的几何变换	cp2tform
Nonreflective similarity	缩放、旋转、剪切和平移的结合；直线保持为直线，平行线保持平行；保持物体的基本形状	cp2tform
Piecewise linear	局部变换的几何变换，不同的仿射变换用于三角形区域	cp2tform
Polynomial	使用2阶、3阶、4阶多项式形式的几何变换	cp2tform
Projective	仿射变换的超集，和仿射变换一样；直线保持为直线，平行线汇聚于尽头	maketform、cp2tform
Similarity	具有额外反射的可能性，类似于非反射相似	cp2tform

7.4 Visualing Aligend Image 观察对准图像

这里自定义一个用于对准图像的函数function visreg。

这个函数的主要功能是将两幅图像根据特征点位置进行覆盖重合，以此来展示对准后的图像。其中，在输入的图片中，一幅是参考图像fref，一般位于对准后的底层；f是输入图像，一般是fref中截取的一部分，用于重合到fref中去。

function h = visreg(fref, f, tform, layer, alpha)
%VISERG Visualize registered images
% VISREG (FREF, F, TFORM) displays two registered images together .

% FREF is the reference image. F is the input image, and TFORM
% defines the geometric transformation that aligns image F with
% image FREF .
% 
% VISREG (FREF, F, TFORM, LAYER) displays F transparently over FREF
% if LAYER is 'top' ; otherwise it displays FREF transparently over
% F.
% 
% VISREG(FREF, F, TFORM, LAYER, ALPHA) uses the scalar value 
% ALPHA, which ranges between 0.0 and 1.0, to control the level of
% transparency of the top image. If ALPHA is 1.0, the top image
% is opaque. If ALPHA is 0.0, the top image is invisible .
% 
% H = VISREG(...) returns a vector of handles to the two di splayed
% image objects. H is in the form [HBOTTOM, HTOP].
if nargin < 5
    alpha = 0.5;
end

if nargin < 4
    layer = 'top';
end

% Transform the input image, f, recording where the result lies in 
% coordinate space
[g, g_xdata, g_ydata] = imtransform(f, tform);

[M, N] = size(fref);
fref_xdata = [1 N];
fref_ydata = [1 M];

if strcmp(layer, 'top')
    % Display the transformed input image above the reference image
    top_image = g;
    top_xdata = g_xdata;
    top_ydata = g_ydata;
    % The transformed input image is likely to have regions of black 
    % pixels because they correspond to "out of bounds" locations on
    % the original image. (See Example 5.2.) These pixe1s should be
    % displayed completely transparently. To compute the appropriate 
    % transparency matrix, we can start with a matrix  filled with the
    % value ALPHA and then transform it with the same transformation
    % applied to the input image. Any zeros in the result will cause
    % the black "out of bounds" pixels in g to be displayed
    % transparently.
    top_alpha = imtransform(alpha * ones(size(f)), tform);
    
    bottom_image = fref;
    bottom_xdata = fref_xdata;
    bottom_ydata = fref_ydata;
else
    % Display the reference image above the transformed input image
    top_image = fref;
    top_xdata = fref_xdata;
    top_ydata = fref_ydata;
    top_alpha = alpha;
    
    bottom_image = g;
    bottom_xdata = g_xdata;
    bottom_ydata = g_ydata;
end

% Display the bottom image at the correct location in coordinate space.
h_botton = imshow(bottom_image, 'XData', bottom_xdata, 'YData', bottom_ydata);
hold on

% Display the top image with the appropriate transparency
h_top = imshow(top_image, 'XData', top_xdata, 'YData', top_ydata);
set(h_top, 'AlphaData', top_alpha);
% The first call to imshow above has the effect of fixing the axis limits.
% Use the axis command to let the axis limits be chosen automatically to
% fully encompass both images.
axis auto

if nargout > 0
    h = [h_botton, h_top];
end

7.5 Area-Based Registration

对于外在特征选择和匹配的可供选择的方式基于区域的配准。在这个过程中，我们是将模版图像的第一幅图像移动到覆盖第二幅图像的每个位置。在每个位置中，计算相似性(similarity metric)。

对于Area-Based Registration，我们需要对similarity metric归一化处理为correlation coefficient相关系数。那么input image于template image之间的归一化互相关定义为：
$$\gamma (x,y)=\frac{{\sum }_{s,t}[w(s,t)-\bar{w}][f(x+s,y+t)-\overline{f_{xy}}]}{\sqrt{{\sum }_{s,t}[w(s,t)-\bar{w}{]}^2[f(x+s,y+t)-\overline{f_{xy}}{]}^2}}$$
其中，$w$是模版，$\bar{w}$是模版元素的均值，只计算一次。$f$是图像，$\overline{f_{xy}}$是$w$和$f$覆盖区域图像的均值。求和是对图像和模版覆盖的$s$和$t$进行。

在MATLAB中，使用function normxcorr2实现：

g = normxcorr2(template, f)

Example Using function normxcorr2 to locate a template in an image：

>> onion   = rgb2gray(imread('onion.png'));
>> peppers = rgb2gray(imread('peppers.png'));
>> imshowpair(peppers,onion,'montage')
>> c = normxcorr2(onion,peppers);
>> figure, surf(c), shading flat
>> [ypeak, xpeak] = find(c==max(c(:)));
>> % Compute translation from max location in correlation matrix
>> yoffSet = ypeak-size(onion,1);
>> xoffSet = xpeak-size(onion,2);
>> figure
>> hAx  = axes;
>> imshow(peppers,'Parent', hAx);
>> drawrectangle(hAx, 'Position', [xoffSet+1, yoffSet+1, size(onion,2), size(onion,1)]);

Example Using normxcorr2 to register two images differing by a translation：

>> g1 = normxcorr2(w, f1);
>> g2 = normxcorr2(w, f2);
>> [y1, x1] = find(g1 == max(g1(:)));
>> [y2, x2] = find(g2 == max(g2(:)));
>> delta_x = x1 - x2;
>> delta_y = y1 - y2;
>> tform = maketform('affine', [1 0 0; 0 1 0; delta_x delta_y 1]);
>> visreg(f1, f2, tform)

至此，我们上面讲到的***7.2***和**7.5都属于手工处理的图像配准方法