Opencv learning -- geometric transformation of image processing

Image zoom

Zoom is to adjust the size of the image , Is to enlarge or reduce the image .

Scaling function ：cv2.resize()

The function prototype ：img=cv2.resize（src,dsize,dst=None,fx=None,fy=None,interpolation=None）

src： original image

dsize： The size of the output image     ( Tuple mode ) [ namely （ wide , high ）]

dst： Output image

fx： Scale factor for scaling along the horizontal axis

fy： Scale factor for scaling along the vertical axis

interpolation： Interpolation method , There are the following 5 Kind of ：

cv2.INTER_NEAREST	Nearest neighbor interpolation
cv2.INTER_LINEAR	Bilinear interpolation （ Default ）
cv2.INTER_AREA	Resampling using pixel region relationship
cv2.INTER_CUBIC	4x4 Double of pixel neighborhood 3 Sub interpolation
cv2.INTER_LANCZOS4	8x8 Pixel neighborhood Lanczos interpolation

If you want to reduce the image , Suggested choice ：cv2.INTER_AREA; If you want to enlarge the image ,cv2.INTER_CUBIC The effect is better but the speed is slow ,cv2.INTER_LINEAR The effect is acceptable and the speed is fast . When zooming , dsize and fx、fy One out of two .

 

 

Image translation

Set the image in the specified direction and distance , Move to the appropriate location .

Translation function ：cv2.warpAffine()

The function prototype :dst=cv2.warpAffine(img,M,dsize)

img： The input image

M：2*3 Move the matrix

about （x,y） Pixels at , To move it to （x+tx,y+ty） Location time ,M The matrix should be set as follows ：

take M Set to np.float32 Type of Numpy Array .

dsize： The size of the output image

The size of the output graph , It should be （ Width , Height ） In the form of .width= Number of columns ,height= Row number .

 

 

Image rotation

The process of rotating an image by a certain angle according to a certain position , The image is still rotating Keep original size . The horizontal axis of symmetry of the image after image rotation 、 The vertical axis of symmetry and the origin of the central coordinate may be transformed , Therefore, it is necessary to convert the coordinates in the image rotation .

stay OpenCV in , Image rotation first obtains the rotation matrix according to the rotation angle and rotation center , Then transform according to the rotation matrix , You can realize the rotation effect of any angle and any Center .

Rotation matrix acquisition function ：cv2.getRotationMatrix2D()

The function prototype ：M=cv2.getRotationMatrix2D（center,angle,scale）

center： Center of rotation

angle： Rotation Angle

scale： Zoom ratio

Get rotation matrix M after , Call again cv2.warpAffine() You can complete the rotation of the image .

Code ：

img = cv.imread('test.jpg')
rows, cols=img.shape[:2]
M=cv.getRotationMatrix2D((cols/2, rows/2), 90, 1)
dst=cv.warpAffine(img,M,(cols, rows))
cv.imshow('xuanzhuan', dst)
cv.waitKey(0)

Affine transformation

The affine transformation of an image involves the change of the shape, position and angle of the image , Affine transformation is mainly the scaling of the image 、 rotate 、 A combination of flip and pan operations .

What is affine transformation ？ Map the corresponding points in the corresponding graph one by one , But the overall shape has changed greatly . Through two groups of three points （ Points of interest ） Find the affine transformation , Affine transformation can be applied to all points in the image , So as to complete the affine transformation of the image .

Create affine transformation matrix function ：cv2.getAffineTransform()

The function prototype ： M=cv2.getAffineTransform(src,dst)

src： original image

dst： Output image

Code ：

img = cv.imread('test.jpg')
rows, cols=img.shape[:2]
pts1 = np.float32([[50, 50], [200, 50], [50, 200]])    # To find three points 
pts2 = np.float32([[100, 100], [200, 50], [100, 250]])
M = cv.getAffineTransform(pts1,pts2)
dst = cv.warpAffine(img,M,(cols,rows))
cv.imshow('fsbh', dst)
cv.waitKey(0)

 

Transmission transformation

Transmission transformation is Change of perspective Result , It's using the perspective center 、 Like point 、 The condition that the target point is three points collinear , According to the law of perspective rotation, make the surface of the image ( Perspective surface ) Around the trace ( Perspective axis ) Rotate an angle , Destroy the original projection beam , The transformation that can still keep the projection geometry on the projection surface unchanged .

 

Create perspective transformation matrix function ：cv2.getPerspectiveTransform()

The function prototype ：T=cv2.getPerspectiveTransform( src,dst)

src： original image

dst： Output image

Image perspective moving function ：cv2.warpPerspective()

The function prototype ：dst=cv2.warpPerspective(img,T,dsize)

img: The input image

T：3*3 Move the matrix

take T Set to np.float32 Type of Numpy Array , from cv2.getPerspectiveTransform() obtain .

dsize： The size of the output image

Code ：

img = cv.imread('test.jpg')
rows, cols=img.shape[:2]
pts1 = np.float32([[56, 65], [368, 52], [28, 387], [389, 390]])    # To find four points 
pts2 = np.float32([[100, 145], [300, 100], [80, 290], [310, 300]])
T = cv.getPerspectiveTransform(pts1, pts2)
dst = cv.warpPerspective(img, T, (cols, rows))
cv.imshow('tsbh', dst)
cv.waitKey(0)

 

Image pyramid

Up sampling function ：cv2.pyrUp()

The function prototype ：dst=cv2.pyrUp(img)

Down sampling function ：cv2.pyrDown()

The function prototype ：dst=cv2.pyrDown(img)

Code ：

img = cv.imread('test.jpg')
up_img = cv.pyrUp(img)
down_img = cv.pyrDown(img)
cv.imshow('up', up_img)
cv.imshow('down', down_img)
cv.waitKey(0)

summary ：