Indexing method for image search engine
First Claim
1. A method of feature vector indexing, comprising the steps of:
- providing a plurality of feature vectors, each feature vector indicative of one of a plurality of objects, wherein each feature vector comprises at least one primitive, each primitive being associated with an attribute of the object and identifying a plurality of features, each feature having an associated feature coefficient;
using a distance metric to select index values which are indicative of features of the feature vector;
providing a target feature vector and a plurality of user weights;
generating a constraint based on the target feature vector and the plurality of user weights; and
applying the constraint to the index values so as to select a subset of the feature vectors.
13 Assignments
0 Petitions
Accused Products
Abstract
One aspect of the invention is directed to a search engine having indexed retrieval to improve computational efficiency of searching large databases of rich objects such as images. Feature vectors are extracted from images and stored in a feature vector database. When a query is submitted to the engine, a query feature vector Q will be specified, as well as a distance threshold T, indicating the maximum distance that is of interest for this query. All images within a distance of T will be identified by the query. Range constraints are defined such that all feature vectors within a distance of T of the query feature vector, satisfy all of the range constraints. The constraint is dependent on the specific primitive being indexed. The constraint is also defined such that any feature vector which is within a distance of T of the query feature vector also satisfies a functional constraint. By reducing the number of feature vectors retrieved from the database and the number of feature vector comparisons, the query process becomes much more efficient.
-
Citations
21 Claims
-
1. A method of feature vector indexing, comprising the steps of:
-
providing a plurality of feature vectors, each feature vector indicative of one of a plurality of objects, wherein each feature vector comprises at least one primitive, each primitive being associated with an attribute of the object and identifying a plurality of features, each feature having an associated feature coefficient; using a distance metric to select index values which are indicative of features of the feature vector; providing a target feature vector and a plurality of user weights; generating a constraint based on the target feature vector and the plurality of user weights; and applying the constraint to the index values so as to select a subset of the feature vectors. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19)
-
-
20. A method of feature vector indexing, comprising the steps of:
-
providing a plurality of feature vectors, each feature vector indicative of one of a plurality of objects, wherein each feature vector comprises at least one primitive, each primitive being associated with an attribute of the object and identifying a plurality of features; using a distance metric to select a plurality of index values which are indicative of features of the feature vector; and generating an index structure corresponding to the feature vectors and accessed by the selected index values.
-
-
21. A method of feature vector indexing, comprising the steps of:
-
providing a plurality of feature vectors, each feature vector indicative of one of a plurality of objects; providing an index structure corresponding to the feature vectors, wherein the index values of the index structure are indicative of features of the feature vectors; providing a target feature vector and a plurality of user weights; generating a constraint based on the target feature vector and the plurality of user weights; and applying the constraint to the index structure so as to select a subset of the feature vectors.
-
Specification