9_20Image_20Processing_20I

Document Sample
9_20Image_20Processing_20I Powered By Docstoc
					                                                                          10/19/2006




                              ENGG 167
             MEDICAL IMAGING
                  Lecture 9: Thursday Oct. 12

                     Image Processing I
              Image Types & Linear Transforms


       Reference: Gonzalez, Woods, Eddins
       Digital Image Processing using MATLAB, Prentice Hall, (2004)   1




Preparation -


Review Imaging Processing Toolbox Help Manual
(on your computer)

Download NIHImage (Mac) / ImageJ (Windows)
(from the web… just do a Google search)


  Most of the next two lectures are adapted from
  Textbook by Gonzalez, Woods, Eddins
  Digital Image Processing using MATLAB, Prentice Hall, (2004)




                                                                      2




                                                                                  1
                                                                                   10/19/2006




Human light perception is sensitive but not linear




                                         Gray scale perception is not linear
                                                                               3
                                                Ref: Gonzalez et al, Text




This leads to ‘optical illusion’ in certain cases




Gray scale perception is not linear




                                      Contrast, length, orientation are not
                                                entirely linear
                                                                               4
                                                Ref: Gonzalez et al, Text




                                                                                           2
                                                                10/19/2006




Image sensors – quantization of light from images




                                                            5
                                Ref: Gonzalez et al, Text




 Grey scale values… humans can discriminate
 about 128 levels (7 bit digitization needed)

     256      128          16                8




     64        32           4                2




                                                            6
                                Ref: Gonzalez et al, Text




                                                                        3
                                                                         10/19/2006




Enhancing images – Intensity transforms

    Negative transform
    Contrast Stretching
    Logarithm transform – gamma factor
    Thresholding (upper and lower)




                                                                     7
                                         Ref: Gonzalez et al, Text




Enhancing images – imadjust transform

Negative transform of image f
g = imadjust(f, [low_in high_in], [low_out high_out], gamma);

g = imadjust(f, [0 1], [1 0]); or g = imcomplement(f);




                                                                     8
                                         Ref: Gonzalez et al, Text




                                                                                 4
                                                                              10/19/2006




       Enhancing images – contrast stretching
                           (Radiology: windowing)
               high I
 Image out
               Low I




                          Low I         high I
                            Image in
                 high I
   Image out
                 Low I




                           Low I         high I
                             Image in
                 high I
   Image out
                 Low I




                           Low I         high I
                             Image in                               9
                                                  Ref: Gonzalez et al, Text




       Enhancing images – gamma factor
               high I
 Image out




                                    gamma = 1
               Low I




                          Low I         high I
                            Image in
             high I




                          gamma > 1
Image out
             Low I




                          Low I         high I

                           Image in
             high I
Image out




                                  gamma < 1
             Low I




                          Low I         high I                     10
                             Image in             Ref: Gonzalez et al, Text




                                                                                      5
                                                                                                 10/19/2006




Enhancing images – Histogram transforms –in
paint shop pro…




                                                               Histogram stretch



                                                                midtone compression


                                                                midtone expansion

                                                                                            11
                                                                   Ref: visible human project




Enhancing images – Histogram transform
(histogram equalization, stretching)
 h(rk) = nk the histogram, h, is the number of pixels, n, at each k intensity
 level, with r being the new scale level ordinate
                                         # of pixels




                                                        0   Grey scale values         255
                                          # of pixels




                                                        0   Grey scale values         255   12
                Ref: visible human project




                                                                                                         6
                                                                                  10/19/2006




    Enhancing images – Histogram processing
                  & resulting transform curves




                                                                             13
                                                 Ref: Gonzalez et al, Text




    Enhancing images – Histogram processing
                  & resulting transform curves
                                                      Repeating again on
                                                      this edge corrects
                                                      this locally




Scanned image ^
(text on either side of
page is showing
through!)                 Removing this
                          large peak at the
                          top of the intensity
                          range eliminates all
                          this ‘crosstalk’




                                                                             14
                                                 Ref: Gonzalez et al, Text




                                                                                          7
                                                                         10/19/2006




Enhancing images – Histogram processing
              global vs local processing




                                                                    15
                                        Ref: Gonzalez et al, Text




Enhancing images – Spatial Filtering &
                   Processing



          1) Linear spatial filtering

          2) Nonlinear spatial filtering

          3) Geometric transformations




                                                                    16
                                        Ref: Gonzalez et al, Text




                                                                                 8
                                                                    10/19/2006




Spatial Filtering with a filter mask to map the
initial image onto a new calculated image




                           Smoothing masks




                                                               17
                                   Ref: Gonzalez et al, Text




Spatial Smoothing Filter masks




                          Increasing mask size “blurs” the
                          image.




                                                               18
                                   Ref: Gonzalez et al, Text




                                                                            9
                                                                10/19/2006




Edge enhancement filter masks - Laplacian




                                                           19
                               Ref: Gonzalez et al, Text




Edge enhancement filter masks – other derivatives




                                                           20
                               Ref: Gonzalez et al, Text




                                                                       10
                                                                                          10/19/2006




Image enhancement via multistage processing




                                                                                     21
                                                        Ref: Gonzalez et al, Text




Assignments this week

Journal Club Presentations: Monday, Oct. 16.
Image processing paper

Image Processing with MATLAB: Due Wed. Oct. 18.
Download assignement.
Choose one Function in MATLAB Image processing toolbox and describe and
demonstrate the function, with an example.
               Options :
               Dilation and Erosion
               Image Deblurring
               Color space transform
               Linear Filter design
               Radon Transform
               Segmentation
               Something else…?
1)   Use an image and show the effect that can be achieved with the MATLAB function.
2)   Show when the effect is dominant and when it is not dominant (what are the main
     control parameters).
3)   Include all MATLAB programs in your handed in work, as well as all images       22




                                                                                                 11

				
DOCUMENT INFO
Shared By:
Tags:
Stats:
views:10
posted:7/9/2010
language:English
pages:11