vsobel − Sobel edge function

vsobel if=infile of=outfile [-i] [-e] [-d] [-f]

Sobel computes the Sobel edge detection function. Isotropic Sobel is optionally specified (which involves sqrt(2) instead of 2 in the weight masks). The use of an Euclidean norm is also optional. The output file format is the same as the input file. The calculated gradients are scaled to be within the range of the input image format. The scale factors are computed from a the highest theoretical response of each type of Sobel operator. The scale factors are as follows :

Non-isotropic w/L1 norm = 1/6

Non-isotropic w/L2 norm = 1/SQRT(20)

Isotropic w/L1 norm = 1/(2 + 2*SQRT(2))

Isotropic w/L2 norm = 1.0/SQRT(8 + 4*SQRT(2))

All numeric image data formats are supported. When the window overlaps the edges of the input image all values beyond the data border are assumed to be zero. For integer and byte images using non-isotropic masks without forced floating point calculations, the L1 norm is always used.

-i

This option selects the isotropic computation, The default is a non-isotropic Sobel mask.

-e

This option selects the Euclidean norm (L2) computation. The default is the sum of absolute differences (L1).

-d

This option is selected then the angle of the gradient is also generated. The result files has two channels: the first is the edge magnitude and the second is the angle. Angle information is scaled to make full use of the 0-255 available range. The four cardinal gradient directions have the following values: East=0, North=64, West=128 and South=192.

-f

This option forces the calculations to be done in floating point. The output format remains that of the input file.

A. P. Reeves