1. Sketchs characterization and compression for nomadic computing
Nelson Baloian, Ramón Cruzat, Richard Ibarra
Departamento de Ciencias de la Computación
Universidad de Chile
Javier Bustos-Jiménez
ORAND

2. Motivation Mobile computing: Learning Design Planning Geology
Topography
Engeneering

3. Our Experience Mobile collaborative applications P2P architechture
Based on sketching & gesturing

4. The problem sketch Sketching originates lots of data
Must be shared in real time
Mobile devices have low memory & processig capacity
sketch

5. The solution Reduce data How ? But this is not new !
Characterizing and “compressing” sketches
But this is not new !
Jones, Durand & Desbrun
Lee et al.
Ramer, Douglas,Peucker
They still demad too much
computation power
Results are not satisfactory for
the mobile scenario

6. Our proposal: Two new algorithms
Distance-based algorithm.
If distance AB is less than e point B can be discarded
Area-based algorithm.
Areas covered by trapezoid ABCX and ACX are compared to decide if B does provide additional information

7. Evaluation (1) Evaluation parameters
Time for characterizing & compressing sketch (t).
Size of resulting characterization (s).
% of information loosing due the compression (l). calculated as ||O − R||/||O||,
O is the pixel matrix of the original image and
R is the pixel matrix of the reconstructed image

8. Evaluation (2) Compare against what ?
Algoritms from the literature already discarted
Strightforward algorothms:
Image algorithm take a photo of the sketch and scale it, for reducing the size of the image.
Compression algorithm use the JavaScript Objection Notation (JSON) for representing the data as an array of pairs [[20.7, 10.2], [8.3, 15.4], ...] and zip it

9. Evaluation tool: X application
Java application for freehand sketching
Transformation algorithms as plug-ins
Generates statistics

10. Statistics: HTML file

11. Experiments - 1 A basic Sketch: rectangle Processed by 4 algoritms
Area with e = 1, 100, 1000 square pixels
Distance with e = 1, 100 and 1000 pixels
Scaling
Coding and zipping

12. Experiments - 2 A basic Sketch: cross Processed by 4 algoritms
Area with e = 1, 100, 1000 square pixels
Distance with e = 1, 100 and 1000 pixels
Scaling
Coding and zipping

13. Experiments - 3 Advanced Sketch: figure Processed by 4 algoritms
Area with e = 1, 100, 1000 square pixels
Distance with e = 1, 100 and 1000 pixels
Scaling
Coding and zipping

14. Experiments - 4 Advanced Sketch: handwriting Processed by 4 algoritms
Area with e = 1, 100, 1000 square pixels
Distance with e = 1, 100 and 1000 pixels
Scaling
Coding and zipping

15. Results: mean values basic sketches
Algorithm
Time [msec]
Mean size [kb] %
Area e=1.0
0.47
0.8
< 1%
Area e=100.0
0.18
0.10
5-6%
Area e=1000.0
0.06
0.03
> 24%
Distance e=1.0
0.35
1.0
0.02%
Distance e=10.0
0.21
0.39
Distance e=100.0
Figure was lost
Scaling to 25%
0.53
18.75
7.5%
ZIP
0.67
0.6
---
RDP Algorithm
2.1
1.39
5 %
Original
3.4 75

16. Mean values for advanced sketches
Algorithms
Time [msec]
Mean size [kb] %
Area e=1.0
0.58
2.2
< 1%
Area e=100.0
0.28
0.35 5%
Area e=1000.0
0.06
0.11
> 40%
Distance e=1.0
0.33
3.5
0.02%
Distance e=10.0
0.23
0.72
0.04%
Distance e=100.0
Figure was lost
Scaling to 25%
0.93
18.75
7.5%
ZIP
0.98
1.74
---
RDP Algorithm
2.4
2.69
20%
Original
3.4 75

17. Subjective evaluation
How does the human user percieves the lose of information

18. Conclusions It does pay off !
Results are valid for the current state of the art -> with more mobile computing power, other algoritms can be evaluated
A methodology for evaluating how good are compression algorithms
Combined with other mechanisms