Acceso usuarios
The visual environment is composed of multiple objects. Some referred to as « artefacts », have non-accidental geometric properties and can be relatively easily described but some others, referred to as natural objects have apparently accidental property with random contours. The contours of artefacts delimits spatially oriented patches wich could be insensitive to spatial high frequency (HF) filtering. The contours of natural objects delimits small elements, such as leaves of a tree, which could be sensitive to H F filtering. Analyzing different types of visual scenes, we found that « landscape » pictures contain more spatial HF than « town » pictures. In a recognition task, recognition rate was higher and response time was lower for « town » pictures than « landscape » pictures. Low-pass filtered and high-pass filtered pictures showed lower performed than full spatial frequency pictures. Recognition performances was equivalent for high-pass filtered « town » pictures and the same low-pass filtered pictures. However, high-pass filtered « landscape » pictures, containing small objects, were less well recognized than the same low-pass filtered pictures. Results suggest two parallel visual processing modes controlled by the relative spatial location of the elements. One involves extracting objects contours when alignment and collinearity are present, the other involves determining patches of equal luminance. The coarse-to-fine scale processing and the continuous-flow models are discussed.
