Visual Representation in the Determination of Saliency

In this paper, we consider the role that visual representation plays in determining the behavior of a generic model of visual salience. There are a variety of different representations that have been employed to form an early visual representation of image structure. The experiments presented demonstrate that the choice of representation has an appreciable effect on the system behavior. The reasons for these differences are discussed, and generalized to implications for vision systems in general. In instances where a design choice is arbitrary, we look to the properties of visual representation in early visual processing in humans for answers. The results as a whole demonstrate the importance of filter choice and highlight some desirable properties of log-Gabor filters.