Granted Patent B2

US 8,768,105 · App. 13/010,807 · Granted Jul 1, 2014

Abstract

Claims (14)

1. A method for searching a database for images that are visually similar to query images, comprising using a processor for

a) extracting visual features from the images in the database;

b) constructing an image anchor graph according to the visual features extracted from the images in the database, wherein the image anchor graph contains a plurality of anchor points corresponding to a subset of images in the database;

c) extracting visual features from the query images;

d) using an inverted file structure, wherein the inverted file structure is a 2D indexing table, for which x-axis corresponds to anchor ID and y-axis corresponds to an image list storing image IDs of images from the database linked to an anchor point; and

e) retrieving images from the database that are similar to one or more query images using a ranking algorithm based on the image anchor graph that compares the visual features extracted from both the query images and the images in the database.

2. The method of claim 1 wherein the ranking algorithm of said step e) is a manifold ranking algorithm.

3. The method of claim 1 , wherein said step b) includes performing K-means clustering on the visual features extracted from the images in the database to obtain a smaller number of anchor points than the number of images in the database, wherein each anchor point is a cluster center.

4. The method of claim 2 , wherein the method further comprises using a processor for learning a ranking function for a query image.

5. The method of claim 2 , further comprising using a processor for:

f) presenting top ranked database images to a user;

g) obtaining feedback from the user about semantic relevance of the top ranked images with respect to the query images;

h) incorporating the user's feedback to re-rank the images in the database; and

i) repeating said steps f) through h) until the user is satisfied.

Assignments (13)