judit tövissy, sándor kopácsi automated stereoscopic 3d image reconstruction for the web dennis...

26
Judit Tövissy, Sándor Judit Tövissy, Sándor Kopácsi Kopácsi Judit Tövissy, Sándor Judit Tövissy, Sándor Kopácsi Kopácsi Automated Automated Stereoscopic Stereoscopic 3D Image 3D Image Reconstruction Reconstruction for for the the Web Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21 2015

Upload: antony-taylor

Post on 22-Dec-2015

229 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Judit Tövissy, Sándor KopácsiJudit Tövissy, Sándor KopácsiJudit Tövissy, Sándor KopácsiJudit Tövissy, Sándor Kopácsi

Automated Automated

StereoscopicStereoscopic

3D Image 3D Image

ReconstructionReconstruction forfor the the

Web Web

Dennis Gabor College, Budapest, HungaryHeraklion, June 18-21 2015

Page 2: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Stereoscopic 3D on the Web

2

• Project started in Hungary, in 2012

• Collaboration between DGC and HAS ICSC

• www.3dweb.hu

Page 3: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

3

Phaistos Disk

disk.3dweb.hu

Page 4: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

4

Phaistos Disk

Page 5: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Distance Representation with Depth Maps

5

Page 6: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

•Issue:– Reverse-engineering of an entire dimension is non-trivial.– Reconstruction needs to be based on agreed upon guidelines

•Assumption #1:– Objects in focus are likely to be closer to the camera than others.

•Assumption #2:– Objects that are brighter are likely to be in the foreground of an

image.

Depth Map Generation for 2D Images

6

Page 7: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Depth Map Generation for 2D Images

7

Page 8: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Resulting 3D Conversion

8

Page 9: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Depth Map Generation for 2D Images

9

Page 10: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Resulting 3D Conversion

10

Page 11: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

StepsStepsofof

ReconstructionReconstruction

11

Page 12: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

• Clustering by Mean Shift Algorithm

Image Segmentation

12

Page 13: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Qualitative Depth Map (QDM)

13

Page 14: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

• Previously Developed by HAS ICSC

Focus Detection

14

Page 15: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

• Combination of a QDM and a Focus Map

Depth-Focus Map

15

Page 16: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

• Based on the previous Depth-Focus Map

Stereoscopic Generation

16

Page 17: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

• 3D Rendering• Raytracing

• Pixel-based 2D Graphics• Von Neumann Neighbours

Innovation: Stencil Filtering

17

• Instead of Von Neumann neighbouring pixels

• Initiate a Recursive Ray in each direction

• Find the first non-blank pixel

Page 18: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

• Recursive Von Neumann Stencil

• Never documented before

• Objective: Find relevant pixels

• Will result in most relevant data

Innovation: Stencil Filtering

18

Page 19: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Stencil Filtering

ResultingPixel

FilteringKernel

RecursiveVon

NeumannNeighbourin

gPixels

OriginalPixel

19

Page 20: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

• Result of an averaging process• Clearly visible edges

Stencil Filtering – Basic Filtering Kernel

20

Page 21: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

• Representation of relevance• Pixels weighted according to distance

Stencil Filtering – Cross Filtering Kernel

21

Bilinear Interpolation

Cross Filtering

Page 22: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Conclusion: Distance is less relevant than was believed

Stencil Filtering – Cross Filtering Kernel

22

Page 23: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Advantage in Realism:Each new pixel is an instance of already existing values in the image

Stencil Filtering – Median Filtering Kernel

23

Page 24: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Stereoscopic 3D Conversion

24

Page 25: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

• S. Battiato, S. Curti, M. La Cascia, M. Tortora and E. Scordato, "Depth map

generation by image classification," Three-Dimensional Image Capture and

Applications VI, pp. 95-104, April 16, 2004.

• L. Kovács and T. Szirányi, "Focus Area Extraction by Blind Deconvolution

for Defining Regions of Interest," Pattern Analysis and Machine

Intelligence, IEEE Transactions on, pp. 1080-1085, June 2007.

• G. Neumann and S. Kopácsi, "Development of 3D Webpages," in CSIT’2013.

Proceedings of the 15th international workshop on computer science and

information technologies, Vienna-Budapest-Bratislava, 2013.

• S. Battiato, A. Capra, S. Curti and M. La Cascia, Writers, 3d Stereoscopic

Image Pairs by Depth-Map Generation. [Performance]. 2004.

Main Bibliographical References

25

Page 26: Judit Tövissy, Sándor Kopácsi Automated Stereoscopic 3D Image Reconstruction for the Web Dennis Gabor College, Budapest, Hungary Heraklion, June 18-21

Judit Tövissy, Sándor KopácsiJudit Tövissy, Sándor KopácsiJudit Tövissy, Sándor KopácsiJudit Tövissy, Sándor Kopácsi

Automated Automated

StereoscopicStereoscopic

3D Image 3D Image

ReconstructionReconstruction forfor the the

Web Web

Dennis Gabor College, Budapest, HungaryHeraklion, June 18-21 2015