A Fast and Efficient VOP Extraction Method Based on Watershed Segmentation

1
A Fast and Efficient VOP Extraction Method Based
on Watershed Segmentation

• Alireza Tavakkoli
• Dr. Shohreh Kasaei
• Gholamreza Amayeh
• Sharif University of Technology

2
Outline

• Problem Definition.
• Literature Review.
• Proposed Method.
• Performance Analysis.
• Conclusions.

3
Content-Based Functionalities in Video

• User Interaction.
• Very Low Bit-rate Video Compression.
• Video Indexing.
• Object-Based Video Coding Systems
• MPEG 4.
• H.263.

4
Video Object Plane (VOP).

• A Video Scene is decomposed into VOPs.
• Multiple VOPs shape Video Object Layers.
• Proposing an Automatic VOP Extractor is a very
  difficult task.

5
VOP Extraction

• Spatio-Temporal Segmentation
• Spatial Segmentation
• Watershed Segmentation.
• Temporal Information
• Motion Detection.
• Change Detection Masks.

6
Watershed Segmentation

• Rain Fall Simulation
• Immersion Simulation
• Vincent Soille (1991).

7
Drawbacks of Watershed Segmentation

• Results in over segmentation.
• Very time consuming.
• Solutions
• Region Merging Techniques.
• Predictive Watershed.

8
Over Segmentation
9
Region Merging Using Modified Gradient
10
Preparing Image for Watershed Segmentation
11
Resulting Watershed Segmentation
12
Predictive Watershed

• Some Definitions

13
Prediction of Watershed
14
Why CDM?

• Distance Learning Application
• Video is supposed to have stationary Background.
• Real time video compression system.
• Stationary Background
• No Global Motion.
• No Camera Motion.
• No Illumination Changes.

15
CDM and Noise

• Difference Noise
• Gaussian Noise
• Due to intensity changes.
• Can be reduced by a Hypothesis test.
• Hypothesis test
• Noise if H0 True.
• Object if H1 True.

16
Region Labeling Method Hypothesis Test
17
Watershed Update
18
VOP Extraction Algorithm
19
Experimental Results
20
Experimental Results (Contd.)
21
Experimental Results (Contd.)
22
Experimental Results (Contd.)
23
Experimental Results (Contd.)
Matlab 6.2 Software, Pentium 4, 2.0GHz.
24
Conclusions

• Ceasing the Over segmentation Problem.
• Speeding up the conventional method using
  temporal information.
• A fast VOP extraction algorithm to be used in
  real time video processing systems.

25
Questions?

• ?