Method, system and apparatus for determining a hash code representing a portion of an image

An apparatus for retrieving feature vectors located within a predetermined search radius, the feature vectors being associated with lattice points of a multidimensional permutohedral lattice, the apparatus comprising:

• means for receiving a query vector extracted from a query image and representing a feature vector to be matched by feature vectors associated with lattice points of the multidimensional permutohedral lattice, the feature vectors being stored in records in a database and representing features of images in the database, the lattice points corresponding to hash values of the feature vectors, wherein the hash values are stored in a memory;
• means for locating, from the memory, a first plurality of lattice points of the multidimensional permutohedral lattice that form an enclosing region around the query vector;
• means for determining a canonical mapping between at least one of the lattice points of the enclosing region and a canonical representation of the enclosing region;
• means for selecting, from the memory, a second plurality of lattice points using the canonical mapping, the second plurality of lattice points being located within the predetermined search radius of a lattice hole of the enclosing region and being distinct to the first plurality of lattice points, wherein the predetermined search radius defines the feature vectors associated with lattice points of the multidimensional permutohedral lattice that are considered a match to the query vector;
• and
• means for retrieving the feature vectors from the first and second plurality of lattice points as matching feature vectors.

A system for retrieving feature vectors located within a predetermined search radius, the feature vectors being associated with lattice points of a multidimensional permutohedral lattice, the system comprising:

• a memory for storing data and a computer program;
• a processor coupled to said memory for executing said computer program, said computer program comprising instructions for:
• receiving a query vector extracted from a query image and representing a feature vector to be matched by feature vectors associated with lattice points of the multidimensional permutohedral lattice, the feature vectors being stored in records in a database and representing features of images in the database, the lattice points corresponding to hash values of the feature vectors, wherein the hash values are stored in the memory;
• locating, from the memory, a first plurality of lattice points of the multidimensional permutohedral lattice that form an enclosing region around the query vector;
• determining a canonical mapping between at least one of the lattice points of the enclosing region and a canonical representation of the enclosing region;
• selecting, from the memory, a second plurality of lattice points using the canonical mapping, the second plurality of lattice points being located within the predetermined search radius of a lattice hole of the enclosing region and being distinct to the first plurality of lattice points, wherein the predetermined search radius defines the feature vectors associated with lattice points of the multidimensional permutohedral lattice that are considered a match to the query vector;
• and
• retrieving the feature vectors from the first and second plurality of lattice points as matching feature vectors.

A non-transitory computer readable medium having a computer program stored thereon for retrieving feature vectors located within a predetermined search radius, the feature vectors being associated with lattice points of a multidimensional permutohedral lattice, the program comprising:

• code for receiving a query vector extracted from a query image and representing a feature vector to be matched by feature vectors associated with lattice points of the multidimensional permutohedral lattice, the feature vectors being stored in records in a database and representing features of images in the database, the lattice points corresponding to hash values of the feature vectors, wherein the hash values are stored in a memory;
• code for locating, from the memory, a first plurality of lattice points of the multidimensional permutohedral lattice that form an enclosing region around the query vector;
• code for determining a canonical mapping between at least one of the lattice points of the enclosing region and a canonical representation of the enclosing region;
• code for selecting, from the memory, a second plurality of lattice points using the canonical mapping, the second plurality of lattice points being located within the predetermined search radius of a lattice hole of the enclosing region and being distinct to the first plurality of lattice points, wherein the predetermined search radius defines the feature vectors associated with lattice points of the multidimensional permutohedral lattice that are considered a match to the query vector;
• and
• code for retrieving the feature vectors from the first and second plurality of lattice points as matching feature vectors.