图像变换

OpenCV中提供的函数

Geometric Image Transformations

官方文档2D变换¹

• 避免采样误差：In fact, to avoid sampling artifacts, the mapping is done in the reverse order, from destination to the source. In case when you specify the forward mapping , the OpenCV functions first compute the corresponding inverse mapping and then use the above formula. 即给定src->dst转换矩阵，实际中也会先算reverse矩阵，再map。
• 原图不存在点的推测（因为即使reverse map，也可能再在原图是float）
  • In the simplest case, the coordinates can be just rounded to the nearest integer coordinates and the corresponding pixel can be used. This is called a nearest-neighbor interpolation. 近邻方法
  • However, a better result can be achieved by using more sophisticated interpolation methods , where a polynomial function. 多项式拟合，具体的方法可以在resize的interpolation参数查到。

getAffineTransform

• 输入两组三个点，计算仿射变换。
• 计算的放射矩阵是2*3矩阵，可以从变换公式中看出，dst的x和src的(x,y)均相关，y亦然。
• 对于点的要求：Given 3 points on one plane（共面很重要） and 3 matching points on another you can calculate affine transform between those planes.
  

getPerspectiveTransform

• 输入两组四个点，计算透视变换。
• 透视变换矩阵3*3.
  

getRotationMatrix2D

• 根据center，angle，scale获取对应的2*3矩阵。
• 上面的变换方法是给定src、dst计算变换矩阵，这个是设计变换矩阵。
  

warpAffine

• 获取了各种2*3的变换矩阵后（比如getAffineTransform、getRotationMatrix2D），用该函数作用于原图，得到变换后的图。
• non in-place

warpPerspective

• 同理，使用3*3的透视变换矩阵，得到变换后的图像。

Motion Analysis and Object Tracking

estimateRigidTransform

• Computes an optimal affine transformation between two 2D point sets. 注意还是2D平面。
• If you can’t get your points with absolute precision (which is normally the case when you get them from an image) then you need more than 3 pairs of points to reduce the error. 如果使用三个点，那么estimateRigidTransform的结果和getAffineTransform结果一致。
• fullAffine参数若为True，则无任何限制共6个自由度；反之则有限制，为5个自由度。

Camera Calibration and 3D Reconstruction

findHomography

• Finds a perspective transformation between two planes.
• This one is more advanced. Not only it can deal with errors in locations of points, but it can also deal with existence of the outliers. 由于estimateRigidTransform使用的是最小化均方误差（i.e 最小二乘值），那么如果有outlier则效果会很差。
• It can use RANSAC or LMeDs to test possible matches and eliminate those outliers.
• It works like（是相似，不是 就是） multiple iterations of estimateRigidTransform: find least square match for different sub-sets of points. If you know that no outliers are present then you can set the ‘method’ argument to 0, and it will work like estimateRigidTransform - by trying to minimize least square error created from match of all points.

其他参考文章²

---

1. Geometric Image Transformations: https://docs.opencv.org/2.4/modules/imgproc/doc/geometric_transformations.html?highlight=getaffinetransform ↩︎

2. 数学求解 http://nghiaho.com/?p=2208 ↩︎