Simple Image Processing Speaker : Lin Hsiu-Ting Date : 2005 / 04 / 27

Outline  Concept of Image Processing  Space Domain Image Processing  Frequency Domain Image Processing  Geometry Transform  Shape Processing  Color System

Concept of Image Processing

Concept  The “Image” Signals We Can See Include Special Information We Process These Signals To Get Relative Information  Integration Technology Engineering Mathematics Physical Biology Medical Science Entertainments

Concept  Application Digital Photo Map Natural Disaster Monitored Others…  Relative Software Photo Shop Photo Impact Others… These Aren’t Today Key Points

Concept  General Topics of Image Process Image Capture & Image Digitize Image Stretch & Remove Distortion Shape Process Image Features Extracted Color Image Process Image Coding & Compression

Concept  Image Digitized Sampling Quantization Coding  Non-Ideal Situations In Process Quantization Error Distortion Noise

Image with Noise  Images Usually Suffer Noise When Sampling (Like Use Scanners or Digital Cameras…)  Some Common Noise Dot Noise Uniform Noise Sinusoid Wave Noise Gaussian Noise Other  Sometimes We Can Remove Noise According Their Features

Image with Noise  Dot Noise  Uniform Noise

Image with Noise  Sinusoid Wave Noise  Gaussian Noise

Space Domain Image Processing

 Characteristic Representation Profile Histogram Statistic ( Mean & Standard Deviation )  Point Operation Binarization Inverse Contract Stretch Histogram Equalization Gamma Correction Arithmetic & Logic Operation

Binarization  Before Binarization ( 8-bit Gray Level )  Binarization (Threshold = 200)

Contract Stretch  Before Processing  After Processing Process Flow Load Image Histogram Statistic Stretch

Histogram Equalization  Before Processing  After Processing Process Flow Load Image Histogram Statistic Equalization

Arithmetic (Add & Sub)  Image #1  Image #2  Image #1 + Image #2  Image #1 - Image #2

Space Domain Image Processing  Range Operation Smoothing ( Low Pass Filter ) Median Filter High Pass Filter Differentiation  Mask Matrix Note ： We Can Also Use 5x5, 7x7 or Larger Matrix Process Range Operation But It Cause More Computing

Median Filter  Before Processing  After Processing For Every 3 x 3 Block Search C n = Median (C) Let f (x, y) = C n Note : The Method Will Have Poor Result When A Lot Of Noise Cluster

Frequency Domain Image Processing

 Fast Fourier Transform  Implement Recursion Algorithm Butterfly Algorithm  Easy To Achieve Filter High Pass / Low Pass Band Pass / Notch

Frequency Domain Image Processing  2D Fast Fourier Transform Do FFT For Every Row …………… Do FFT For Every Column F ( u, v ) Note : We Always Use Log Unit Present The Spectrum Distribute Instead of Linear Because Its Dynamic Range is Larger Then Screen

Frequency Domain Image Processing  Image  Spectrum  Image with Sin Noise  Spectrum

Geometry Transform

 Coordinates Transform Rotation Scaling Twist  Gray Level Interpolation Replicative Interpolation Bilinear Interpolation

Coordinates Transform  Rotation  Scaling  Twist

Gray Level Interpolation  When We Transform From R to R* Some Point In R* Can’t Correspond From R Rotation, Magnify Suffer This Question Ex: Magnify ?2?3? ?????? 4?5?6? ?????? 7?8?9? ??????

Gray Level Interpolation  Replicative Interpolation Use The Nearest Point To Present Let j = Int(x+0.5), k = Int(y+0.5) => g ( x’, y’ ) = f ( j, k )  Bilinear Interpolation Use Four Neighborhood Points More Smooth Than Replicative

Gray Level Interpolation  Replicative Interpolation  Bilinear Interpolation

Shape Processing

 Find The Edges And Bones Binarization  Process The Edge And Bone Erosion Dilation Open / Close Remove Isolate Points  Usually Simple Logic Operation

Erosion & Dilation  Binarization Image  Erosion  Dilation

Color System

 The Colors We See Wave Length 380 nm ~ 780 nm Use Rods to Recognize Brightness Use Cones to Recognize Colors (Three Types For R. G. B. Colors) Usually Eyes Are More Sensitive To Brightness Than Colors This Feature is Convenient For Image Compressing

Color System  Common Color System R. G. B. System (Red, Green and Blue) C. M. Y. System (Cyan, Magenta and Yellow) -- A Complement of R. G. B Y. U. V System Y. I. Q System H. S. I. System

Conclusion  Image Processing Is Useful  Image Processing Is Interesting  Although We Needn’t Know The Details Of Techniques Because Many Powerful Software Will Handle Them…  But Knowing General Concept Is Helpful For Us

Reference  數位影像處理 - 連國珍 著, 儒林出版  /dip/Chapter02.pps 