Systems and methods for image-feature-based recognition

• US 9,122,706 B1
• Filed: 02/10/2015
• Issued: 09/01/2015
• Est. Priority Date: 02/10/2014
• Status: Active Grant

• Current Assignee: Geenee GmbH
• Original Assignee: Geenee Ug
• Inventors: Priehn, Hannes, Holzschneider, Philipp, Magers, Thorsten Otto Wolfgang
• Primary Examiner(s): Koziol, Stephen R
• Assistant Examiner(s): SCHWARTZ, RAPHAEL M
• Application Number:
  US14/619,045
• Publication Number:
  US 20150227557A1
• Time in Patent Office: 203 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/51 Indexing; Data structures therefor; Storage structures
• CPC Class Codes: G06F 16/5838 using colour
• CPC Class Codes: G06F 18/213 Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06T 11/60 Editing figures and text; Combining figures or text
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• CPC Class Codes: G06V 20/20 in augmented reality scenes

Abstract

Methods and systems are described herein that allow a user to capture a single image snapshot from video, print, or the world around him or her, and obtain additional information relating to the media itself or items of interest displayed in the snapshot. A fingerprint of the snapshot is used as a query and transmitted to the server. Image Feature-Based Recognition, as described herein, uses a feature index to identify a smaller set of candidate matches from a larger database of images based on the fingerprint. Novel methods and systems using a distance metric and a radical hash table design exploit probabilistic effects and allow distinct image features to be preferred over redundant ones, allowing only the more distinctive data points to remain resident within the index, yielding a lean index that can be quickly used in the identification process.

Claims

1. A server for identifying and transmitting to a mobile device additional information relating to a query image obtained by a user with a mobile device, the server comprising:

a reference image database storing one or more reference feature descriptors associated with one or more reference images, wherein at least one reference image feature descriptor is mapped to a plurality of first addresses within an address space in the reference image database using a plurality of functions, and wherein the plurality of functions is configured to map one or more feature descriptors to the same address space; and

a processor for executing one or more modules comprising;

a query receiving module for receiving, from the mobile device, a first query including one or more query image feature descriptors;

a candidate feature descriptors retrieval module for;

calculating, for each of the one or more query image feature descriptors, a plurality of second addresses using the plurality of functions, and

for each of the plurality of second addresses, determining a number of duplicate reference image feature descriptors stored within a pre-defined region of the reference image database;

a candidate images retrieval module for;

selecting a set of reference image feature descriptors based on the determined number of duplicate reference image feature descriptors;

an image matching module for;

comparing the selected set of reference image feature descriptors with the one or more query image feature descriptors to determine if any of the one or more reference images match the query image;

wherein, if a matching reference image is determined, the server is further configured to perform the steps of;

obtaining additional information associated with the at least one matching reference image from an annotation database, and

transmitting the additional information to the mobile device for displaying as additional information relating to the at least one item of interest to the user.

2. The server of claim 1, wherein the processor is further configured to execute a hash table management module for:

providing, for each of the plurality of second addresses, one or more references to one or more of the reference image feature descriptors based on a relationship between each of the plurality of second addresses and each of the plurality of the first addresses;

and wherein the number of duplicate reference image feature descriptors is determined based on the relationship.

3. The server of claim 2, wherein the hash table management module further comprises a collision resolving module for:

using a random function to calculate the plurality of the first addresses.

8. The server of claim 1, wherein each of the plurality of first addresses is associated with a single reference image feature descriptor.

9. The server of claim 2, wherein the hash table management module further comprises an overflowing handling module for:

determining whether the one of the plurality of first addresses is associated with a first reference image feature descriptor when the one of the plurality of first addresses is calculated;

responsive to determining that the one of the plurality of first addresses is associated with the first reference image feature descriptor;

disassociating the one of the plurality of first addresses from the first reference image feature descriptor; and

associating the one of the plurality of first addresses with a second reference image feature descriptor.

10. The server of claim 1, wherein the processor is further configured to execute the candidate images retrieval module for:

ranking one or more sets of reference image descriptors, each set being associated with a combination of the determined numbers of duplicate reference image feature descriptors; and

selecting the set of reference image feature descriptors based on a result of the ranking.

11. The server of claim 1, wherein the processor is further configured to execute the candidate images retrieval module for:

excluding, from the selected set of reference image feature descriptors, one or more reference images associated with a calculated number of duplicate reference image feature descriptors that is below a predefined threshold.

12. The server of claim 1, wherein the reference image descriptors are stored within the databased based on a hash table including a plurality of hash buckets, each hash bucket being associated with an address within the address space;

wherein the plurality of functions are configured to map one or more feature descriptors to one or more hash buckets out of the same set of hash buckets.

4. A computer-implemented method for identifying and displaying additional information relating to a first query image obtained by a user with a mobile device, the method comprising:

receiving, from the mobile device, a first query including one or more query image feature descriptors corresponding to at least one item of interest to the user included in the first query image;

responsive to receiving the first query, performing a similarity search, wherein the similarity search comprises;

for each of the query image feature descriptors, calculating a plurality of addresses using a plurality of functions;

for each of the plurality of addresses, determining a number of duplicate reference image feature descriptors stored within a certain region of a reference image database storing one or more reference image feature descriptors, each of the reference image feature descriptors being associated with at least one reference image, wherein at least one reference image feature descriptor is mapped to a plurality of addresses within an address space in the reference image database using the plurality of functions, and wherein the plurality of functions is configured to map one or more feature descriptors to the same address space; and

determining one or more candidate reference images based on a number of duplicate reference image feature descriptors within the certain region of the reference image database associated with each of the plurality of addresses;

comparing the feature descriptors for the one or more candidate reference images to the query image feature descriptors to determine if there is a match; and

if there is a match,

acquiring additional information associated with the match; and

transmitting, to the mobile device, the additional information associated with the match for displaying to the user.

5. The computer-implemented method of claim 4,

wherein the number of duplicate reference image feature descriptors within a pre-defined region associated with an address reflects a degree of similarity between the duplicate reference image feature descriptors and the first query image feature descriptors mapped to the address;

and

wherein the determination of one or more candidate reference images comprises;

determining that the number of duplicate reference image feature descriptors exceeds a pre-determined threshold; and

acquiring the one or more candidate reference images associated with the duplicate reference image feature descriptors.

6. The computer-implemented method of claim 5, further comprising:

receiving, from the mobile device, a second query including one or more second query image feature descriptors;

comparing the one or more first query image feature descriptors with the one or more second query image feature descriptors; and

based on a result of the comparison;

determining not to perform the similarity search; and

acquiring, based on a mapping between the first query and the one or more candidate reference images stored in a query cache, the one or more candidate images.

7. The computer-implemented method of claim 4, wherein

the reference image feature descriptors are stored within a database based on a hash table including a plurality of hash buckets, each hash bucket being associated with an address,

each hash bucket being associated with an address within the address space;

wherein the plurality of functions are configured to map one or more feature descriptors to one or more hash buckets of out the same set of hash buckets.

13. The computer-implemented method of claim 4, further comprising:

receiving, from the mobile device, a second query including one or more second query image feature descriptors;

comparing the one or more first query image feature descriptors with the one or more second query image feature descriptors; and

based on a result of the comparison;

determining not to perform a similarity search; and

acquiring, based on a mapping between the first query and the one or more candidate reference images stored in a query cache, the one or more candidate images.