Image processing device and recording medium storing image processing program

• US 8,660,361 B2
• Filed: 12/02/2011
• Issued: 02/25/2014
• Est. Priority Date: 12/03/2010
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Yu, Shanshan, Nakashima, Satoshi
• Primary Examiner(s): COUSO, JOSE L
• Application Number:
  US13/309,619
• Publication Number:
  US 20120141033A1
• Time in Patent Office: 816 Days
• Field of Search: 382/103, 382/115, 382117-118, 382/154, 382/181, 382/190
• US Class Current: 382/195
• CPC Class Codes: G06V 40/165 using facial parts and geometric relationships

Abstract

An image processing device includes a storage unit that stores dictionary information; a generating unit that extracts, from an input image, a plurality of characteristic point candidates, and generates a plurality of combinations that each include a plurality of characteristic point candidates; a removing unit that removes, for each of the combinations, at least one characteristic point candidate based on at least one of the dictionary information and information obtained by analyzing the input image; and a determining unit that acquires, for each of the combinations, results of matching the dictionary information with the combination of which the at least one characteristic point candidate has been removed, selects a combination of characteristic point candidates based on the acquired matching results, and determines, as the characteristic points, the plurality of characteristic point candidates included in the selected combination.

Claims

1. An image processing device comprising:

a processor;

a storage; and

a memory which stores a plurality of instructions, which when executed by the processor, cause the processor to execute,

storing dictionary information of a combination of a plurality of authentic characteristic points in the storage;

extracting, from an input image, a plurality of characteristic point candidates that are candidates for characteristic points;

generating a plurality of combinations that each include a plurality of characteristic point candidates that are among the extracted characteristic point candidates, the generating generates a plurality of exclusive combinations that each include a single characteristic point candidate for face parts;

removing, for each of the combinations of the characteristic point candidates, at least one characteristic point candidate from the plurality of characteristic point candidates included in the combination based on at least one of the dictionary information and information obtained by analyzing the input image; and

acquiring, for each of the combinations of the characteristic point candidates, results of matching the dictionary information with the combination of which the at least one characteristic point candidate has been removed by the removing from the characteristic point candidates included in the combination, selecting a combination of characteristic point candidates from among the combinations of the characteristic point candidates based on the acquired matching results, and determining, as the characteristic points, the plurality of characteristic point candidates included in the selected combination.

2. The image processing device according to claim 1,

wherein the removing performs, for each of the combinations of the characteristic point candidates, matching of a combination of the plurality of authentic characteristic points with the combination of the characteristic point candidates and removes a characteristic point candidate that causes a larger deviation of distances between the characteristic point candidate and the other characteristic point candidates from distances between the authentic characteristic points, as compared to the other characteristic point candidates.

3. The image processing device according to claim 1,

wherein the removing removes, for each of the combinations of the characteristic point candidates, at least one characteristic point candidate based on likelihoods of the characteristic point candidates for the characteristic points and/or information of a characteristic point candidate removed by a previous process, while the likelihoods are obtained by analyzing a current frame that is the input image.

4. The image processing device according to claim 3,

wherein the removing adjusts the likelihoods based on information of an object'"'"'s orientation obtained by analyzing the current frame and a frame preceding the current frame.

5. A non-transitory computer-readable recording medium having an image processing program recorded thereon, that causes a computer to execute operations, the operations comprising:

extracting, from an input image, a plurality of characteristic point candidates that are candidates for characteristic points;

generating a plurality of combinations that each include a plurality of characteristic point candidates that are among the extracted characteristic point candidates, the generating generates a plurality of exclusive combinations that each include a single characteristic point candidate for face parts;

removing, for each of the combinations of the characteristic point candidates, at least one characteristic point candidate from the plurality of characteristic point candidates included in the combination based on at least one of dictionary information of a combination of a plurality of authentic characteristic points and information obtained by analyzing the input image;

acquiring, for each of the combinations of the characteristic point candidates, results of matching of the dictionary information with the combination of which the at least one characteristic point candidate has been removed from the characteristic point candidates included in the combination; and

selecting a combination of characteristic point candidates from among the combinations of the characteristic point candidates based on the matching results, and determining, as the characteristic points, the characteristic point candidates included in the selected combination.

6. The non-transitory computer-readable medium according to claim 5,

wherein in the removing of the at least one characteristic point candidate, matching of the combination of the plurality of authentic characteristic points with each of the combinations of the characteristic point candidates is performed, a characteristic point candidate that causes a larger deviation of distances between the characteristic point candidate and the other characteristic point candidates from distances between the authentic characteristic points, as compared to the other characteristic point candidates, is removed from the combination.

7. The non-transitory computer-readable medium according to claim 5,

wherein in the removing of at least one characteristic point candidate, at least one characteristic point candidate is removed for each of the combinations of the characteristic point candidates based on the basis of likelihoods of the characteristic point candidates for the characteristic points and/or information of a characteristic point candidate removed by a previous process, while the likelihoods are obtained by analyzing a current frame that is the input image.

8. The non-transitory computer-readable medium according to claim 7,

wherein in the removing of the at least one characteristic point candidate, the likelihoods are adjusted based on information of an object'"'"'s orientation obtained by analyzing the current frame and a frame preceding the current frame.

9. An image processing method comprising:

extracting, from an input image, a plurality of characteristic point candidates that are candidates for characteristic points;

generating a plurality of combinations that each include a plurality of characteristic point candidates that are among the extracted characteristic point candidates, the generating generates a plurality of exclusive combinations that each include a single characteristic point candidate for face parts;

removing, for each of the combinations of the characteristic point candidates, at least one characteristic point candidate from the plurality of characteristic point candidates included in the combination based on at least one of dictionary information of a combination of a plurality of authentic characteristic points and information obtained by analyzing the input image;

acquiring, for each of the combinations of the characteristic point candidates, results of matching of the dictionary information with the combination of which the at least one characteristic point candidate has been removed from the characteristic point candidates included in the combination; and

selecting, by a computer processor, a combination of characteristic point candidates from among the combinations of the characteristic point candidates based on the matching results, and determining, as the characteristic points, the characteristic point candidates included in the selected combination.