Automatic segmentation using multiple cues classification
In this thesis, we proposed a segmentation scheme. We assume that the background has no significant motion and the foreground has some form of motion. Optical flow velocity are gouped into 2 regions, each region is fitted with a Gaussian distribution. A Baysian decision rule [1] is used to decide t...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Theses and Dissertations |
| Published: |
2008
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/4509 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Summary: | In this thesis, we proposed a segmentation scheme.
We assume that the background has no significant motion and the foreground has some form of motion. Optical flow velocity are gouped into 2 regions, each region is fitted with a Gaussian distribution. A Baysian decision rule [1] is used to decide the region a pixel belowngs to. Suitable weights are then assigned to these regions. K-means clustering are used to classify regions in terms of intensity and spatial locations. The percentage of moving pixels is then calculated. Bayesian rule is used to decide which of these regions belongs to foreground or background. fuzzy classification is used to combined the two cues, namely, optical flow velocity and percentage pixel moving into foreground and background. Last but not least, a covolution averaging filter are applied respectively to the resultant mask to remove Gaussian noise. Our proposal method aims to solve the unevenness in the edge which most segmentation methods suffer from. |
|---|