Document Sample
image_processing_2 Powered By Docstoc
					                      Basic Introduction to
                       Image Processing

This presentation has been put together as a common effort of Urs Ziegler, Anne Greet Bittermann, Mathias
Hoechli. Many pages are copied from Internet web pages or from presentations given by Leica, Zeiss and
other companies. Please browse the internet to learn interactively all about optics. For questions &
registration please contact .
Presentation of multidimensional data

3D-data has to be presented in in a 2D-fashion for publication
on paper. The data set might be represented i.e. as image
gallery, top+side view, projection. A virtual light source and/or
shadows on a virtual projection plane are helping to recognize
spatial relations.

Interactive models, movies and animations can be “published”
on web-pages or into power point-presentations.
     Image processing and analysis
After registration on the microscope the digital images are loaded to image
processing software for further processing. The data includes information about
pseudo color, pixel dimensions, time scale etc.

First image data get adjusted by background subtraction, contrast enhancement,
etc. Colors might be assigned; subvolumes selected; z-mismatchs corrected by

The softwares offer different options to look at the multidimensional data sets.
i.e. slice viewer, gallery view, section view, projections, full 3D volume
representations, surface models, time bar, color coded overlays of several
channels, transparencies, ...

The software offers analytical tools for measurement and quantification: automated
counting of features, measurements of areas and volumes, tracing of filaments,
measuring of distances, evaluation of colocalization, ...
   Automated Multidimensional
       Data Processing

Dimensions:   micrograph processing softwares:

xy    = 2D          Imaris (Bitplane)*
xyz = 3D           Volocity (Improvision)
xyzt = 4D              NIH image **
xyzt! = 5D            BioImageXD **
                  •Campus-Lizenz an der Uni Zürich
                  **scientific freeware on the internet
Digital images:
2-dimensional distribution of image Points (Pixel)


Digital resolution
Detectors record a limited amount of image points (pixel number)
within a xy grid. Each image point has its own grey level (dynamic

Increasing the amount of image points as well as the number of
grey levels leads to bigger image files and longer calculation

256 grey levels are coded by 8 bit. 256 grey levels are presented
by a computer monitor.

Today, detectors are pushed to discriminate 1024, 4096 or more
grey levels. The human eye can discriminate about 60 gray levels
(6 bit).
         3D Data set
y                       )!z
     z                  )!z

    The information within
    the optical sections along
    the z-axis can be used to
    reconstruct a 3-dimensional
                       4D Data set

                 3-D stacks recorded along the time course

y                         )!z                                )!z
                          )!z                                )!z
                          )!z                                )!z
    z                     )!z                                )!z

            t1                                    t2
                           5D Data set
Wavelenghts adding another dimension of fluorescent data. Time laps of
multi-channel 3D stacks generate a 5D data set. Wavelenght information
is displayed as pseudo-colors.
   y                         )!z                     )!z                 )!z
                             )!z                     )!z                 )!z
                             )!z                     )!z                 )!z
       z                     )!z                     )!z                 )!z

                                   t1, t2, t3, ...
A voxel (= volume element) is the 3D-equivalent of the
2D-pixel. It is the smallest unit of a sampled volume.

 The given maximal lateral (x,y) resolution of 0.2 µm
 and the axial (z) resolution of 0.4 µm of a voxel
 results in an elongated shape (point spread function).

For the calculation and visualisation are the neighbor voxels
of great importance.

         2D                                         3D
  -> each Pixel has                           -> each Voxel has
  4 neighbor pixels                           6 neighbor voxels
Presentation and effort:

* Simple presentations (fast, allows 2D-publishing):
gallery view, section view, projections

* Intense calculations (time consuming, for analysis):
full 3D volume representation, surface rendering,
shadowing, stereo view

* Animations (time consuming, analysis & presentation):
rotating 3D models, time sequences of 3D
Image Gallery

                Galleries of
                images are the
                most simple
                data presen-

                for xyz
                    xy! ...
Projecting optical sections to one plane

                              Optical section
                              through a cube
                              containing fibers

                     Projecting the structures
                     of all sections to the ground
                     level („Extended Focus“)
Projection types

Average Projection:
Simple to very complex mathematical procedures. Summing up the grey
values of all voxels with identical xy-coordinates along the z-stack,
divided by the numbers of optical sections.

Maximal Intensity Projection (MIP):
Only the voxel in the z-stack, which has the highest grey value, will be

Background signal gets projected too and might cause noise/blur.
Suppress background first!!
                      x1                                       x1
                 y1                   Z1                  y1                 Z1

                      x2                                       x2
                 y2                   Z2                  y2                 Z2

                           x3                                       x3
            y3                        Z3                 y3                  Z3

                                      Z4                                     Z4

                                      Z5                                     Z5

Maximal                                      Averaging               Projektion
Intensity                                    may lead to
Point                                        enlarged
Projection                                   structures
-> sharp image                               and background
    Gallery presentation of a neuron

1         2         3         4

5         6         7         8

9         10
    of the
    of the
    with one
            stack of images

                                         gallery of images
                       Section through
y   z                     the stack
                                          Image of the section


Sectioning through
a stack of images
- perpendicular
                       y    z



                     Section through
                     the stack along
                     the y-axis
                    X-Y   Y-Z

of section levels
   in XY, XZ, YZ

Intense calculation
for 3-D representations
1. Volume rendering
      Ray tracing

2. Surface rendering
      Segmentation of z-stacks
      Depth encoding of voxels

3. Animations
      time course
      zooms etc.
Volume rendering

Even if fog (background) limits the visibility, we get an
idea of the structure of the trees.
Volume rendering

                                 Volume                       Screen

   Virtual ray

Ray Tracing
A virtual ray passing the volume accumulates the grey levels of the
voxels, normalizes the summed value and presents it on the screen.
Volume rendering with adjustments of the grey values

Adjustment of the grey
level according to the
distance between voxel
and screen.

                  Voxels hit by the virtual ray

                                          Adjustment of the grey
                                          value according to the grey
                                          value of the voxel just
                                          passed.                       screen
Volume rendering - example

  3D representation of a multifluorsescent cellmonolayer
                       (4 channels)
 Surface rendering

Creating objects with solid surfaces.
Surface rendering: Iso-Surface modelling
1st Step: Segmentation of the z-stacks. Identification of Voxels
belonging to an object.The criteria for the identification is the grey
value of the voxel.

All voxels, whose grey value are higher (brighter) than the chosen
threshold belong to the object, the others belong to the background
and will be discriminated. This threshold value is chosen by the

(Neighborhood rule: If a voxel belongs to the
object, but one of its 6 neighbor voxels does
not belong to the object, it will be defined as a
surface voxel.)
Surface rendering: Iso-Surface modelling
2nd Step: Depth encoding of the Voxels.

The previously identified surface voxels have all the same grey
value and would result as a non structured evenly grey image on the
screen of the monitor. Therefore, in a second step, the grey values of
the voxels are adjusted according to the distance of the surface voxels
to the screen.
                                 distance (depth)

                                                         Depth dependent
                 All voxels have the                adjustment of the grey
                 same grey value                                   values.
  Surface rendering: Iso-Surface modelling
  3rd step: Shadowing

The topology can be accentuated using a one sided shadowing effect.
To do that, neighboring surface voxels are connected to form a polygon.
The grey values of the surface voxels are adjusted dependent on the an-
gle between the viewing direction and the normal of the polygon surface.

     Viewing direction
     and incident light
                          The normal
                !         to the poly-
                          gon and the
  Sur-                    direction
  face voxels             include the
  define polygons          angle " .
Representation of several surfaces

                                  e) + Transparency
Th resholds 110 (red) und 60 (whit
  Surface modeling: setting the threshhold

Threshold 68                          Threshold 138

 Surface models of the same dendrites using different threshold values
               Which model shows the real surface ?
Adequate Filament Imaging
The depth feeling can be simulated by calculating two separate slightly tilted
3D-models of the same scene as if they were viewed by the left eye and the right
eye. The final stereo pair can be observed using different techniques.

The 3D impression can be achieved squinting the eyes or using special stereo
viewers (or crossing the eyes).
   ation II

  The 2 pictures
    of the stereo
   are colored in
red & green and

     The 3D im-
pression can be
 achieved using
bicolor goggels.
Looking inside

           Surface view

           Surface view combined
           with the visualization of
              internal structures
Gallery view of 20 optical sections                       Section view of 20 sections
                                                                x-y             y-z


Mo l,stained with acridine orange - 20 optical sections    3D-representation: x-y, x-z, y-z
              Looking inside

Transparency & slicer tool
Looking inside

       ... by using transparency
       Animations - fly through

Volume and surface
rendering allow you to
turn and zoom the data
set. Extreme Zoom
allows you to virtually
enter the sample.

- Automated data segmentation
- Particle counting
- Size regognition
- Distance measuerments
- Filament tracking
- Movement tracing

" Results are visualized in the 3D model
" Results are listed as numbers in Exel-sheets

The relation of the intensity values
from 2 channels are presented in
a two dimensional histogram.

In case of colocalization, the inten-
sity clouds of both channels are

Colocalization is not an absolute
fact but allways relate to voxel size
and resolution.
Animations are series of single images put together into a movie. The images might be a
volume view, a projection, a slice, a time point. The animation is done by just playing the
sequential data set, or by rotating 3D models or volume representations, by zoom-in & fly-
through motions, changing of surfaces and transparencies, etc.

                                          Today#s computer allow to calculate and
                                          represent animated sequences reasonably
                                          fast. Movie files can be published i.e in
                                          power point or on the web. Also interactive
                                          file formats are possible.
     Animation in time



     Changes of a 3D-volume with time might be
     presented as a gallery of projection views -
t4   or as a movie. Animation and stereo view
     facilitate the recognition of spheric relations
     in this context.

and tracing
in time
What is to be gained?

• Increase in resolution x, y, z
• Noise is reduced
• The image formation process is optimized
  (astigmatism, point spread function, ...)
Widefield fluorescent data can be improved a lot by decon-

Confocal data show less z-distortions, less out-of-focus blur,...
-> deconvolution shows only very little effect.
Convolution - Theory

   Fluorescent bead with a diameter of 0,1 µm
Deconvolution procedure
            •Measure object of known size, but smaller than the
            resolution of the microscope (i.e. 100nm fluorescent

            •Compare the microscope image with the
            ideal/theoretical representation of the object.

            •Determine the difference of the measured and the
            real object.

            •Correct unknown objects with the determined
Deconvolution effect
3D, 4D, 5D- data reconstruction is time consuming!!!

=>Only correctly recorded images are worth to spend the time to deal
  with the 3D presentation!!!

=> Keep your data small:
 ° Reduce image resolution (512 x 512 pixel = 262 kB).
 ° Crop images so that they containing only the most important structural details.
 ° Work with as less channels as possible.
 ° stay with 8 bit

=>Keep the coffee pot hot in order to wait patiently until the calculations are finished.

=>Use classical image processing tools to improve the quality of the images.

and: Don$t expect to much of a 3-D presentation.