Scale selection using three different representations for images

Abstract Humans are capable of zooming in on the right range of scale, but it is not clear, yet, whether this processing involves either spatial variables, or two-dimensional spatial-frequency variables, or both spatial and spatial-frequency variables in the image description. The main motivation of this work is to show that at least it is possible in each domain to formulate algorithms giving solutions to this problem of scale selection. We also propose that in situations where no information is available about appropriate scales for analysis, a reasonable approach is to consider descriptions at the spatial scales for active sensors under a data-driven multisensor organization, with the sensors being orientation and spatial-frequency selective. Of course, the sensor scales are defined as those of patterns with frequency components in the sensor.