We present a video segmentation algorithm that blends color based region segmentation into feature-based motion estimation in an iterative clustering framework. The algorithm does not require any user assistance to accurately extract object boundaries from color video. After filtering and simplification, marker points are selected to grow 3D volumes around them by evaluating local color attributes. The volumes are refined and motion trajectories are obtained. Relational properties of volume pairs are captured in the form of mutual descriptors that are computed from motion trajectories. These descriptors designed to characterize shape as well as spatial properties of volumes. The trajectories are then used to initialize feature-based motion estimation. In the clustering stage, volumes are merged into objects iteratively until a motion similarity score of merged objects becomes negligible. Finally, a multi-resolution object tree that gives the video object planes for every possible number of objects is generated.
[1]
Murat Kunt,et al.
Second-generation image coding
,
2000
.
[2]
Edward H. Adelson,et al.
Representing moving images with layers
,
1994,
IEEE Trans. Image Process..
[3]
Patrick Bouthemy,et al.
Motion segmentation and qualitative dynamic scene analysis from an image sequence
,
1993,
International Journal of Computer Vision.
[4]
Serge Beucher,et al.
Watershed, Hierarchical Segmentation and Waterfall Algorithm
,
1994,
ISMM.
[5]
J.K. Aggarwal,et al.
Correspondence processes in dynamic scene analysis
,
1981,
Proceedings of the IEEE.
[6]
Patrick Bouthemy,et al.
Region-Based Tracking Using Affine Motion Models in Long Image Sequences
,
1994
.
[7]
Manfred Kopp,et al.
Efficient 3 x 3 Median Filter Computations
,
1994
.
[8]
Josef Bigün,et al.
Spatio-Temporal Robust Motion Estimation and Segmentation
,
1995,
CAIP.