Generic Shape Incorporation for Probabilistic SpatioTemporal Video Object Segmentation Rakib Ahmed,

1
Generic Shape Incorporation for Probabilistic
Spatio-Temporal Video Object SegmentationRakib
Ahmed, Gour C. Karmakar and Laurence S. Dooley
Gippsland School of Information Technology,
Monash University, Victoria 3842, Australia.
Abstract
Recently a probabilistic spatio-temporal video
object segmentation algorithm incorporating shape
information (PST-S) has been proposed, though
since it is providing higher priority to pixels
near the cluster centre, which theoretically
limits the segmentation performance of the
technique. To address this problem, a new
probabilistic spatio-temporal video object
segmentation algorithm that incorporates generic
shape information based on its region (PST-RS)
has been proposed. Experimental results reveal a
significant performance improvement in
arbitrary-shaped video object segmentation
compared with other contemporary methods.
Key Contributions
Motivation

• Integration of region based generic shape
  information in the GMM model by combining the
  concepts of uniform distribution and confidence
  interval of Gaussian distribution.
• Incorporation of probabilistic shape information
  which impacts on the likelihood of pixels being
  labelled or assigned to a particular cluster.

• The most important perceptual attribute in
  distinguishing and recognising objects is shape.
• As an object is perceptually represented by a
  region within its shape, incorporation of
  region-based shape information will improve video
  object segmentation performance.

Experimental Results
The density function of the distribution of a
mixture of k Gaussians for a Random variable
is
The maximum likelihood (ML) estimation of
for a set is
Frame 16
Frame 16
The shape prior function for a pixel
belonging to layer j of frame t
Representation of shape using an object region
Pixel labeling
(b) PST-S
(a) PST
Frame 48
(c) PST-RS
Frame 69
(c) PST-RS
(a) PST
(b) PST-S
Figure 1
Segmentation of Carphone sequence
Segmentation of Carphone sequence
Conclusion

• New video object segmentation technique
  seamlessly incorporating region-based generic
  object shape information into a PST segmentation
  framework.
• Improving the overall segmentation performance in
  comparison with PST and PST-S techniques.
• No increase in the order of computational
  complexity than PST technique.

Videos and images have been taken from the
Internet
Contacts Rakib.Ahmed, Gour.Karmakar,
Laurence.Dooley_at_infotech.monash.edu.au