Combining soft decision algorithms and scale-sequential hypotheses pruning for object recognition
- Northeastern Univ., Boston, MA (United States)
This paper describes a system that exploits the synergy of Hierarchical Mixture Density (HMD) estimation with multiresolution decomposition based hypothesis pruning to perform efficiently joint segmentation and labeling of partially occluded objects in images. First we present the overall structure of the HMD estimation algorithm in the form of a recurrent neural network which generates the posterior probabilities of the various hypotheses associated with the image. Then in order to reduce the large memory and computation requirement we propose a hypothesis pruning scheme making use of the orthonormal discrete wavelet transform for dimensionality reduction. We provide an intuitive justification for the validity of this scheme and present experimental results and performance analysis on real and synthetic images to verify our claims.
- OSTI ID:
- 466456
- Report Number(s):
- CONF-9610138-; TRN: 97:001309-0036
- Resource Relation:
- Conference: International multi-disciplinary conference on intelligent systems: a semiotic perspective, Gaithersburg, MD (United States), 21-23 Oct 1996; Other Information: PBD: 1996; Related Information: Is Part Of Intelligent systems: A semiotic perspective. Volume I: Theoretical semiotics; Albus, J.; Meystel, A.; Quintero, R.; PB: 303 p.
- Country of Publication:
- United States
- Language:
- English
Similar Records
Block-classified bidirectional motion compensation scheme for wavelet-decomposed digital video
Wavelets as basis functions to represent the coarse-graining potential in multiscale coarse graining approach