Shape representation is fundamental to many areas of computer vision. Local symmetries provide a tool on which to base flexible representations of complex shapes, however, 'multiple participation' symmetries such as SLS generate many more symmetries, than required. This clutters the representation with mostly irrelevant symmetries, hiding perceptually salient information. In this paper, a salience hierarchy is proposed that allows compact, perceptually pleasing representations to be extracted from the highly redundant full SLS. Both individual and competitive saliency is used to create the hierarchy which partitions the set of symmetries into subsets, each of which is an independent shape representation.
- Local symmetries
- Shape representation
ASJC Scopus subject areas
- Signal Processing
- Computer Vision and Pattern Recognition