Data enhancement in image processing

Data enhancement is mainly divided into Spatial domain enhancement And Frequency domain enhancement
Spatial threshold enhancement only changes the gray level of this point , Do not change its position .

One 、 linear transformation

（1）s=255-r（ Image reversal ）

（2）s = Ar + B

（3） Piecewise linear transformation

Two 、 Nonlinear transformation

（1） Logarithmic transformation

For darker images

（2） Power transformation

r >1, For brighter pictures
r <1, For darker pictures

3、 ... and 、 Algebraic operations

（1） Addition operation

Definition ： Add two or more pictures
application ： Remove additive noise

（2） Subtraction

Definition ： Subtract the two pictures
application ： Split specific areas ; Detect scene changes

（3） Multiplication

Definition ： Multiply the two images
application ： Get a specific part of the image

Make one mask A mask

Four 、 Spatial domain filtering

Mean filter （ Convolution reduces noise ）
Gaussian low pass filter

application

Because the mean filter will blur the edge , Therefore, median filter appears

Sort multiple values , Median value

• Make the highlight bright （ dark ） spot , Closer to the point around it
• Eliminate isolated brightness or dark spots
• While removing noise , Keep the edges better
• It can effectively remove impulse noise （ Black and white dots are superimposed on the image ）
  

High pass filtering ： Sharpen the image

• Highlight the details of the image
• Enhance the blurred edges of the image
  
  
  
  
• The first derivative can detect whether a pixel in the image is on the edge
• The second derivative can judge whether an edge pixel is on the bright or dark side
  

5、 ... and 、 spectrum shaping

** The principle of frequency domain filtering ：** Convolution in spatial domain corresponds to the product of frequency domain .

• Enhance based on spatial domain , That is to convolute the input image with the image enhancement system .
• Frequency domain based enhancement , The frequency domain expression of the input image is obtained by Fourier transform , Then the desired components are obtained through the frequency domain enhancement system （ Unwanted ingredients ）, Then the enhanced image can be obtained by inverse Fourier transform .

H(u, v) Determines the effect of enhancement .

（1） low pass filter , Remove （ attenuation ） High frequency components , Sharp details in the image are smoothed .

• Ideal low pass filter
  

There is a ringing phenomenon ：

• Butterworth low pass filter （ Slow decay ）
  

• Gaussian low pass filter
  
  
  application ： Remove grid effects 、 Face wrinkle removal 、 Text fracture repair

（2） High pass filter , Remove （ attenuation ） Low frequency components , Edges in the image are enhanced .

• Ideal high pass filter

There is a ringing phenomenon

• Butterworth High pass filter , There is a ringing phenomenon
• Gaussian high pass filter , No ringing

Mixed use of enhancement methods ：

（3） Homomorphic filtering

low pass filter 、 High pass filter can only solve Additive noise problem , Cannot be reduced Multiplicative or convolutional noise , So homomorphic filtering appears .

• Enhance contrast
• Compress the gray dynamic range in the frequency domain
• Based on image imaging model （ Here's the picture ）
  
  

Fourier transform the logarithmic transform of the imaging model


After doing inverse Fourier transform and exponential transform, the output image can be obtained .

application

（4） be based on Retinex wave filtering

application