VLOGGEN | VisionX V4 |
vloggen − generates blob-like candidates from an image using multiscale Laplacian of Gaussian (LOG) filtering
SYNOPSISvloggen [if=<inputimage>] [ig=<inputmask>] [of=<outputfile>] [og=<prefix>] [d1=<mindiameter>] [d1=<maxdiameter>] [n=<n>] [k=<dogk>] [-t] [-v]
DESCRIPTIONvloggen detects blob-like structures on the <inputimage> using multiscale LOG filtering. The algorithm uses <n> characteristic sizes increasing from <mindiameter> to <maxdiameter> in geometric progression. For each size, scale-normalized LOG kernel is applied to the <inputimage> resulting in a set of filtered images that form four-dimensional image feature space (three spatial dimensions + characteristic size). Then all local maxima within spatial <inputmask> are found. Spatial locations of these maxima along with corresponding characteristic size make up the list of possible blob candidates in <outputfile>.
The convolution is implemented in Fourier domain, with scale-normalized LOG kernel substituted by DOG kernel with parameter <dogk> for which closed-form transform is known. Transforms of DOG kernels depend on the image resolution and are computed in runtime.
OPTIONS|
if= |
input image file in floating point format | ||
|
ig= |
search space mask image file in byte format; default: the entire image is used for search | ||
|
of= |
output candidate file, containing blob candidates; the columns are: candidate index, x,y,z coordinates, diameter (in mm), and response of the LOG filter | ||
|
og= |
prefix for outputting filtered images at each scale (floating point format); default: no image output | ||
|
d1= |
smallest charactristic size, i.e. smallest diameter of the blob candidate to be detected, default 3; (for the images with DICOM history, the measurement unit is mm) | ||
|
d2= |
largest charactristic size, default: 25 | ||
|
n= |
number of charactristic sizes, default: 10 | ||
|
k= |
k parameter of the DOG kernel: sigma2 = k * sigma1, default: 1.01 | ||
|
-t |
use tab-separated format instead of default VisionX feature format both for input and output candidate files; input file may have arbitrary number of columns, however the first three must be the following: candidate index, x,y,z coordinates and diameter (in mm) | ||
|
-v |
verbose mode; gives some runtime information |
* only accepts floating point 3D images not greater than 512x512x512 in dimensions. If a dimension of the input image is not a power of two, the image is padded with zeros along this dimension prior to the convolution.
* required amount of RAM for a generic CT scan (512x512xZ):: 2048MB + 3 * size of <inputimage> + (size of <inputmask>, if specified)
SEE ALSOvloggenw(1), vlogtype(1), vlogratio(1), vattratio(1), vairdist(1), vpsdist(1), vcurvature(1), vcompact(1)
AUTHORSergei Fotin, August 2008