mt3000/3dsift-rank

3DSIFT-Rank

FeatExtract

featExtract is the program used to extract SIFT-Ranked [1] features from images. It accepts nifti images (.nii, .hdr, .nii.gz) as input, and output a list of keypoints and their descriptors.

Usage

Volumetric local feature extraction v1.1
Usage: featExtract [options] <input image> <output features>
      <input image>: nifti (.nii,.hdr,.nii.gz).
      <output features>: output file with features.

Options

-w	Output feature geometry in world coordinates, NIFTI qto_xyz matrix (default is voxel units).
-2+	Double input image size.
-2-	Halve input image size.

Output

The program will output a .key file (with the same name as the input file), containing a list of features, with their coordinates, scale, orientation, and descriptor.

FeatMatchMultiple

featMatchMultiple is the program, based on FLANN library [2], used to match features.

Usage

Volumetric Feature matching v1.1
Usage: featMatchMultiple [options] -f <input filelist>
      <input filelist>: Text file containing the list of .key files (from featExtract). 1 key file per line.

Option

-n	Number of nearest neighbors. Default: 5.

Output

The program will output different files:

_command.txt	The complete command used to generate this results (for logging purpose).
_names.txt	List of the input image filename
matching_votes.txt	NxN matrix containing the accumulation of the weighted votes for each pair of image. The most important file.
feature_count.txt	The number of features for each input file.
vote_count.txt	NxN matrix containing the accumulation of the non-weighted votes for each pair of image.

Data

https://central.xnat.org/data/projects/SIFTFeatures

References

[1] Toews, Matthew, and William Wells. "Sift-rank: Ordinal description for invariant feature correspondence." 2009 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2009.
[2] Muja, Marius, and David G. Lowe. "Scalable nearest neighbor algorithms for high dimensional data." IEEE transactions on pattern analysis and machine intelligence 36.11 (2014): 2227-2240.

Contact

[email protected], [email protected]