1. MATCHSLIDE : INT contribution
Patrick HORAIN
Hichem ATTI
Waheb LARBI
Presented as : "TELESLIDE: Technical aspects ", Jacques Klossa & Patrick Horain,
Joint Congress of Quantitative Analytical Cellular Pathology ans Telepathology , September , Heraclion, Crete.

2. Wide fields : some issues
Acquisition :
A microscope can see only a small field in a slide
Browsing wide fields :
Narrow band networks
Narrow computer screens

3. Wide field acquisition

4. Mosaicing
Grab neighbour fields images and merge (mosaic) them into a wide field image
Ex.: 16 images, 1300  1030 pixels
1 image, 4662  3675 pixels
Shading !
Mosaicing = assembling individual slide parts into a single image or wide field.
Problem: illumination inhomogeneities

5. Shading : what ? Problems : Example empty slide :
Illumination inhomogeneities
Dust in optical system
Example empty slide :
original
stretched histogram

6. Shading correction / single field
Normalise each pixel with illumination / =

7. Shading correction / wide field
With shading correction
Without shading correction
Comparer les 2 images par des allers-et-retours au clavier (touches page suivante / précédente).

8. Other issues Future work includes : color correction
robust slides registration for mosaicing
hierarchical RoI definition by image analysis

9. Wide field browsing

10. Browsing requirements
Narrow band networks
Use better compression
Use progressive transmission
Use regions of interest
Narrow computer screens
Use interactive windowing while loading
 Use the new JPEG 2000 standard

11. What is JPEG 2000 ? New ISO Image compression Standard Wavelet-based
Lossless or lossy compression
Better low bit-rate performance than JPEG
Progressive transmission by pixel accuracy and resolution
Random code stream access and processing
Interactive zooming and regions of interest

12. JPEG 2000 : wavelets Discrete wavelets transform (DWT)
recursive image decomposition into frequency subbands
At each level of the DWT, the low frequency component (upper left) is subsampled by a factor 2 in each direction.

13. JPEG 2000 versus JPEG JPEG2000 (0.099 bpp) Original (24 bpp)
DWT integrates a spatial and frequency representation of images without the blocks effect of DCT.
At low bit rates, less image degradation with JPEG 2000 than with JPEG.
JPEG2000 (0.099 bpp)
Original (24 bpp)
JPEG (0.18 bpp)
DCT used in JPEG involves a decomposition of images on a base of cosine functions with infinite support. Keeping the details local in images is achieved by splitting images to small blocks, which introduces artefacts visible at high compression.

14. Progressive transmission : resolution levels
Resolution 5 : 1300  1030 pixels (100 % of image data)
Resolution 4 : 650  515 pixels (25 % of image data)
Resolution 3 : 325  258 pixels (6.25% of image data)
Resolution 2 : 163  129 pixels (1.61 % of data)
Resolution 1 : 82  65 pixels (0.45% data)

15. Progressive transmission : quality layers
Progression on the precision of the DWT coefficients
1rst layer : 12 kb (7.36 % of image data)
15th layer : 27 kb (16.6 % of image data)
20th layer : 163 kb (100 % of image data)
5th layer : 14.9 kb (9.14 % of image data)
10th layer : 18.1 kb (11.1 % of image data)
Optimised rate allocation control on code blocks

16. Progressive transmission : other modes
Progression modes :
R : Resolution (DWT levels)
L : Quality Layers
C : Components
P: Position
5 std progression orders (apply from right to left) :
LRCP, RLCP, RPCL, PCRL, CPRL

17. The difference between the progression orders can be perceived only at very high compression rates (< 1 bit per pixel).
LRCP is the default progression order :
for each L // quality layer
for each R // resolution
for each C // component
for each P // position along lines & columns
In the 2 rightmost columns, progression on components (luminance first, then chrominance) is apparent.

18. Regions of interest (RoI)
RoI can be coded with different fidelity
ex.: lossless RoI vs. lossy background
RoI = a rectangle or an arbitrary mask
RoI data transmitted before background

19. RoI as a rectangle

20. RoI as an arbitrary mask

21. Progressive transmission with RoI
0.399 bpp
1.133 bpp

22. Other issues Future work includes :
comparative evaluation of the progression modes w. r. to our target application
automatic RoI selection by image analysis
compression and interaction on multiple focus images