Scale selection for differential operators

The scale problem is common to most tasks involving analysis of image data—in general situations it is hardly ever possible to know in advance at what scales interesting structures can be expected to appear. Size variations of image structures can occur for several reasons, e.g. because the world contains objects of different physical size, because surface textures contain structures at different scales, and because of perspective effects and noise in the image formation process. A proper representation of scale is essential to most visual tasks requiring stable descriptors of image structure. In certain problems, such as shape-from-texture, scale variations in an image also constitute a primary cue in its own right.