数字图像处理期末知识点总结（英文版）

Digital image processing Final exam summaries

1. Brightness adaptation(亮适应)
2. Lens--晶状体  cones--锥状体—bright-light vision  rod--杆状体—dim-light vision
3. Sampling(取样)  quantization(量化)—uniform and non-uniform
4. Neighbor:4-neighbors、D-neighbors、8-neighbors
5. Connectivity:4-connectivities、m-connectivities、8-connectivities
6. Distance metric:
   1. educildean distance- 
   2. 4-distance(city diatance)-
   3. 8-distance(chessboard distance)-
7. Histogram(直方图)—histogram equalization、histogram specification

      Gray imageàintensity level histogram(many to one)

      Horizontal:intensity level  verital:number of pixel

1. PDF:probility ditensity function
2. Spatial resolution(空间分辨率) and intensity resolution(灰度分辨率)
3. Smoothing spatial filter(blur the edge):
   1. Mean filter:linear
   2. Lowpass filter:ideal lowpass filter、butterworth lowpass filter、gaussian lowpass filter
   3. Order-statistic filter:nonlinear
      1. median filter：local，nonlinear，used to eliminate impulse(salt&peppey) noise
      2. maximum filter
      3. minimum filter
      4. midpoint filter
      5. alpha-trimmed mean filter(修正的阿尔法均值滤波器)
4. Sharpening spatial filter(high-light the transition in intensity):
   1. Highpass filter:ideal highpass filter、butterworth highpass filter、gaussian highpass filter
   2. Gradient:

      

First-order diverative,peoduce thicker edges,be used to detect the presence of edge in an image

1. 1. Laplacian:

Second-order diverative,has strong responses to detail,has a transition from positive to negative,be used to determine the exact position of edges

1. 1. Both of c) and d) are used to detect edge and sensitive to noise
   2. Laplacian sharpening results in larger intensity discontinuity near the edge
2. Smoothing and sharpening
   1. Similarity:both are used to enhance image and improve the quality of image
   2. Difference:smoothing filter can eliminate noise,but also easily causes the egde blur;sharpening filter can enhance edge,resulting in high-frequency component enhancement,make the image clear
3. 2-dimensional discrete fourier function:

1. Homomorphic filtering
   1. Reason:(1)An image can be expressed as the peoduct of its illumination and reflectance.(2)The illumination conpoment is associated with the low-frequencies in the image,while the reflectance conpoment is associated with the high-frequencies.
   2. Step:f(x,y)→in→DFT→H(u,v)→DFT-1→export→g(x,y)

      

   3. Goal:dynamic range compression and contrast enhancement at the same time.
2. Image restored is an inverse process of image degradation
3. Image enhancement and image restored
   1. Similarity:(1)both can be used to improve image quality.(2)both can be used in spatial or frequency domain.
   2. Difference:(1)image enhancement is a subjective process,while image restored is an objective process.(2)the result of image enhancement is determined by the perception of the viewer,while the result of image restored is determined by a criterion.
4. Noise:Gaussian,Impulse(salt&peppey),Exponential,Uniform,Rayleigh,Gamma,periodic noise
5. Ringing effect
   1. Ideal lowpass filter>butterworth lowpass filter,gaussian lowpass filter doesn’t have ringing effect
   2. Reason:the loss information in the process of image degradation,especially the loss of high-frequency-ripples
   3. Result:It seriously reduces the quality of the image restored and makes it difficult to follow up the subsequent steps of image restored.
6. Laplacian operator in frequency domain:

1. Filter
   1. Mean filter:arithmetic mean filter,geometric mean filter,harmonic mean filter,inverse harmonic mean filter
   2. Order-statistic filter:median filter,maximum filter,minimum filter,midpoint filter,alpha-trimmed mean filter
   3. Adaptive filter:
      1. adaptive local noise reduction filter:want to preserve the edge
      2. adaptive median filter:nonlinear,want to eliminate impulse noise while preserving the edges
   4. Wiener filter(minimum mean square error filter):usually be used to restore image,the noise power spectrum and image power spectrum must be calculated.
2. Coloring image
   1. RGB:red,green,blue—monitor,color camera
   2. HIS:hue,situation,intensity—closely correspond with human’s eye
   3. CMY(K):cyan(青色),magenta(深红色),yellow(,black)—color painting
   4. For RGB color image enhancement,it can use weighted mean filter.
3. Pseudo coloring and false coloring
   1. Difference:pseudo coloring is the process of coloring a gray image to a color image by a criterion;false coloring is the process of coloring a image to get a different color image from the natural image.
   2. Similarity:both want to represent an image by a color sensitive way by human’s eye.
4. Image compression
   1. Reason:(1)spatial redundancy (2)interpixel redundancy (3)psychovisual redundancy
   2. Goal:reduce the amount of data that required to represent a digital image
5. Watermark:

   

6. Lossless compression and lossy compression
   1. Lossless:can completely restore the original data without causing any distortion
      1. Huffman coding
      2. Run-length coding
      3. Arithmetic coding
      4. LZW coding
   2. Lossy:predict coding,based-symbol coding,transform coding,wavelet coding
7. Wavelet transform and Fouriver transform:wavelet transform can achieve orthogonalization and remain time domain characteristics,but its frequency characteristics is poorer than Fourier transform.
8. Up-sampling:enlarge the image

Down-sampling:shrink the image 

1. Dilation: enlarge the image  

   

Erosion:shrink the image    

1. Opening operator:

   

        Closing operator:

A∙B=(A⊕B)⊝B

     2.Boundary detection:

        Holes filling:

1. Top-hat:

   

        Bottom-hat:

        Top-hat operator can correct the effect of non-uniform illumination

        Use both of them can enhance the contast.

1. Image segmention
   1. Point detect:Laplacian mask,give high response to the object that has the similar shape as the mask such as isolated point.
   2. Line detect:four shape masks-verital,horizontal,diagonal(45 degree and -45 degree)
   3. Edge detect:use highpass filter and will enhance noise
2. LoG operator and DoG
   1. LoG:Laplacian of gaussian
   2. DoG:difference of gaussian
   3. Similarity:
      1. both can be realized directly by using one-dimensional diverative instead of two-dimensional diverative
      2. the first step of them is smoothing the image
   4. Difference:LoG requires more addition and multiplication than LoG.
3. Step of edge detect:
   1. Smooth the image to reduce noise
   2. Detect the edge points
   3. Edge localization
4. Canny operator
   1. Goal:
      1. Low error rate
      2. Edge should be well localized
      3. The detector should return only one point for each true edge point
   2. Step:
      1. Smoothing the input image with a gaussian filter
      2. Compute the gradient magnitude and angle images
      3. Apply nonmaximal suppression to the gradient magnitude
      4. Use double thresholding and connectivity analysis to detect and link edge.
   3. Nonmaximal suppression
      1. Find the direction dk that is closest to a(x,y)
      2. If the value of m(x,y) is less than at least one of its two neighbors along dk,set gn=0,else,set gn=m(x,y)
   4. Double thresholding
      1. Locate the next unvisited edge pixel p in gnh
      2. Mark as valid edge pixels all the weak pixels in gnl that are connect to p using 8-connectivity
      3. If all nonzereo pixels in gnh have been visited,go to next step,else,return to step one
      4. Set to zero all pixels in gnl that were not marked as valid edge pixels