HOG = VL_HOG(IM, CELLSIZE) computes the HOG features for image IM and the specified CELLSIZE. IM can be either grayscale or colour in SINGLE storage class. HOG is an array of cells: its number of columns is approximately the number of columns of IM divided by CELLSIZE and the same for the number of rows. The third dimension spans the feature components.
PERM = VL_HOG('permutation') returns the left-right permutation to apply to each HOG cell to flip it.
IMAGE = VL_HOG('render', HOG) returns an IMAGE containing an iconic representation of the array of cells HOG.
Options:
- Variant 'UoCTTI'
Choose a HOG variant: 'UoCTTI' or 'DalalTriggs'.
- NumOrientations 9
Choose a number of undirected orientations in the orientation histograms. The angle [0,pi) is divided in to NumOrientation equal parts.
- DirectedPolarField
By specifying this flag the image IM is interpreted as samples from a 2D vector field specified by their argument IM(:,:,2) and modulus IM(:,:,1).
- UndirectedPolarField
Same as above, but wraps angles in [0,pi).
- BilinearOrientations
This flags activates the use of bilinear interpolation to assign orientations to bins. This produces a smoother feature, but is not some other implementations (e.g. UoCTTI).
- Example computing and visualizing HOG features
hog = vl_hog(im2single(im)) ; % compute HOG features
See also: HOG fundamentals, VL_HELP().