The calculation of salient structures is one of the early and basic ideas of perceptual organization in Computer Vision. Saliency algorithms aim to find image curves, maximizing some deterministic quality measure which grows with the length of the curve, its smoothness, and its continuity. This note proposes a modified saliency estimation mechanism, which is based on probabilistically specified grouping cues and on length estimation. In the context of the proposed method, the wellknown saliency mechanism, proposed by Shaashua and Ullman [SU88], may be interpreted as a process trying to detect the curve with maximal expected length.
The new characterization of saliency using probabilistic cues is conceptually built on considering the curve starting at a feature point, and estimating the distribution of the length of this curve, iteratively. Different saliencies, like the expected length, may be specified as different functions of this distribution. There is no need however to actually propagate the distributions during the iterative process.
The proposed saliency characterization is associated with several advantages: First, unlike previous approaches, the search for the “best group” is based on a probabilistic characterization, which may be derived and verified from typical images, rather than on pre-conceived opinion about the nature of figure subsets. Therefore, it is expected also to be more reliable. Second, the probabilistic saliency is more abstract and thus more generic than the common geometric formulations. Therefore, it lends itself to different realizations of saliencies based on different cues, in a systematic rigorous way. To demonstrate that, we created, as instances of the general approach, a saliency process which is based on grey level similarity but still preserve a similar meaning. Finally, the proposed approach gives another interpretation for the measure than makes one curve a winner, which may often be more intuitive to grasp, especially as the saliency levels has a clear meaning of say, expected curve length.
- Feature Point
- Length Distribution
- Human Visual System
- Perceptual Organization
- Probabilistic Interpretation
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.