1.Scanner - Geometric transformation

Image filtering

1. Transformation of image and shear transition

• Meaning of geometric transformation
  • The task to change the whole image by altering arrangement of pixel composing image.
  • ex) Image registration, removal of geometric distortion, etc.
    

• Translation transformation
  • The transformation moving the image as specific size to horizontal or vertical direction.
  • It have to set the translation displacement of x-axis and y-axis

• Affine transformation function

cv2.warpAffine(src, M, dsize, dst=None, flags=None, borderMode=None, borderValue=None) -> dst

• src : Input image
• M : 2 x 3 affine metrix
• dsize : Result image size
• dst : Output image
• flags : Interpolation type
• borderMode : Edge pixel expension method
• borderValue : default value is 0

• Example of translation transformation of the image

• Shear transformation
  • It have to set each x-axis and y-axis.

2. Shrinkage and enlargement of image

• Scale transformation
  • The transformation that size of image make larger or smaller than original image.
  • Specifying the scale factor of x-axis and y-axis

• Scale transformation code

cv2.resize(src, dsize, dst=None, fx=None, fy=None, interpolation=None) -> dst

• src : Input image
• dsize : Output image size
• dst : Out image
• fx,fy : scale factor of x,y direction
• interpolation : decide interpolation

• Scale transformation example

• Scale transformaion precaution
  • When scale transformation be applied to image, Detail of image disappear
  • After Filtering smoothly input image, It apply to step by step shirinkage.
  • cv2.INTER_AREA flag is used in cv2.resize() function of OpenCV

3. Image pyramid

• Image pyramid
  • Image sets of various resolution for one of image
  • Generally, It’s consist of shrinkage size using Gaussian blurring & Down sampleing

• Image pyramid downsampling code

cv2.pyrDown(src, dst=None, dstsize=None, borderType=None) -> dst

• src : Input image
• dst : Output image
• dstsize : Output image size
• borderType : Edge pixel expansion method *** note
  • Firstly, It’s applied to Gaussian filter of 5x5 size. Then, Small size of image is made from delete even number of column and row.

• Image pyramid upsmapling code

cv.pyrUp(src, dst=None, dstsize=None, borderType=None) -> dst

• src : Input image
• dst : Output image
• dstsize : Output image size
• borderType : Edge pixel expansion method

• Image pyramid example

4. Rotaion of image

• Rotation transformation
  • Transformating image to rotate by specific angle

• Rotation transformation code

cv2.getRotationMatrix2D(center, angle, scale) -> retval

• center : Rotational center coordination
• angle : Rotaion degree
• scale : Additional magnification ratio
• retval : 2 X 3 Affine transformation matrix

5. Affine Tansform and Perspective Transform

• Diffrent point between affine transform and perspectiv transform

• To get affine transform matrix

cv2.getAffineTransfrom(src, dst) -> retval

• src : Three point of original coordinate
• dst : Three point of result coordinate
• retval : 2 x 3 perspective transform matrix

cv2.getPerspectiveTransfrom(src, dst) -> retval

• src : Four point of original coordinate
• dst : Four point of result coordinate
• retval : 3 x 3 perspective transform matrix

• Perspective transform example

6. Remapping

• Remapping
  • Normal process specific position pixel of image rearranes to other position

• Remapping could express various trasformation with affine transform, perspective transform

• Remapping code

cv2.remap(src, map1, map2, interpolation, dst=None, borderMode=None, borderValue=None) -> dst

• src : Input image
• map1 : x-coordination of input image
• map2 : y-coordination of output image
• interpolation : Interpolation
• dst : Output image
• borderMode : Edge pixel expansion method
• borderValue : Constant value

• Remapping example using trigonometic function