This method was first elaborated by KUMPETI [Kumpeti, 1985], then improved by KHATIR [Khatir, 1988]. The principle is the following:

• The original image is first processed by a SOBEL operator to get a binary image of edge points.
• The binary image is then reduced by mosaic, each point of the reduced image counts the number of edge points of a window 8x8 or 16x16 of the binary image.
• The reduced image is then segmented by means of a hysteresis thresholding, to keep only the strongly textured areas.

The method is fast thanks to the reduction of data, but the edges of the urban areas are coarse. The use of 8x8 windows gives better defined edges, but small areas tend to be kept whereas they are not built-up areas.

The method proposed by M. ROUX [Roux, 1992] is inspired by the concept of edge points density and uses tools from mathematical morphology:

• The original image is first processed to bring the clear peaks and the dark peaks to the same grey level, by means of a morphological difference: a closing and an opening of ray 1 are subtracted.
• The image thus obtained is reduced by average on 4x4 windows.
• Remaining linear structures are discarded by an opening of ray 1.
• This result is then made binary to provide a mask in order to select pixels in the original image.