Each texture measure calculated will be stored as width-height planes along the elements direction in the Output Object (o). The order in which the (selected) texture measures will be stored in the output object are:
L5L5 L5E5 L5S5 L5W5 L5R5 E5L5 E5E5 E5S5 E5W5 E5R5 S5L5 S5E5 S5S5 S5W5 S5R5 W5L5 W5E5 W5S5 W5W5 W5R5 R5L5 R5E5 R5S5 R5W5 R5R5
If the input object elements dimension is greater than 1 then the results obtained by applying the first selected kernel to each band are stored by planes in the output object, followed by the results obtained by applying the next selected texture kernel to all bands, and so on. Thus a 3-band input image, operated on by the L5L5 and R5R5 kernels, will produce a 6-band output image with the results as follows:
Band 0 - L5L5 on Band 0 of input image Band 1 - L5L5 on Band 1 of input image Band 2 - L5L5 on Band 2 of input image Band 3 - R5R5 on Band 0 of input image Band 4 - R5R5 on Band 1 of input image Band 5 - R5R5 on Band 2 of input image
The five center-weighted vectors are:
L5 = [ 1 4 6 4 1] E5 = [-1 -2 0 2 1] S5 = [-1 0 2 0 -1] W5 = [-1 2 0 -2 1] R5 = [ 1 -4 6 -4 1]
Each 5 x 5 kernel is derived from multiplying a particular combination of two of the above vectors. This results in 25 possible 5 x 5 kernels. Note that 10 of the kernels are formed by taking the transpose. For example, L5E5 is a 5 x 5 kernel formed by multiplying the vectors, L5 with E5. It's transpose is E5L5, and is similarly formed by multiplying the vectors, E5 with L5. One of these kernels is more sensitive to horizontal changes in texture, while the other is sensitive to vertical changes in texture.
The kernels are applied with a centered hotspot.
The sum of the elements of each kernel is zero, which results in the output image having a mean of zero. Therefore the relevant texture information is contained in the image variance of the microtexture features. The LAW kernels were designed to be sensitive to structures such as edges, ripples, and spots.
Choosing which kernel to use will depend on the nature of the texture of interest, and will require trial and error to decide which produces the feature with the most discriminating power. In general, if a texture kernel of say, L5S5 is used, one may also want to use its transpose, S5L5 for certain types of quasiperiodic variations commonly found in textured images.
The colorspace model for the output object is always set to KNONE.
Group; specify AT LEAST ONE of:
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
AND/OR
The output object will be of type KLONG, KDOUBLE, or KDCOMPLEX, determined by the data type of the input object. If the input object is of data type KBYTE, KUBYTE, KSHORT, or KUSHORT, then it is converted up to an KLONG image. If the input is of type KFLOAT or KDOUBLE, then the result will be KDOUBLE. Any complex input type is converted to type KDCOMPLEX.
J. Y. Hsiao, and A. A. Sawchuk, "Supervised textured image segmentation using feature smoothing and probabilistic relaxation techniques" IEEE Trans. Pattern Anal. Machine Intell., vol. 11, No. 12, pp 1279-1292, 1989.