Object identification device, method, and storage medium

• US 9,633,278 B2
• Filed: 12/25/2013
• Issued: 04/25/2017
• Est. Priority Date: 12/28/2012
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Mase, Ryota
• Primary Examiner(s): Li, Ruiping
• Application Number:
  US14/655,836
• Publication Number:
  US 20150356374A1
• Time in Patent Office: 1,217 Days
• Field of Search: 382195
• CPC Class Codes: G06T 2207/10004 Still image; Photographic image
• CPC Class Codes: G06T 7/60 Analysis of geometric attributes
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]

Abstract

Disclosed is an object identification device and the like for reducing identification error in a reference image which presents an object that is only slightly difference from an object presented in an input image. The object identification device includes a local feature quantity matching unit for calculating geometric transformation information for transformation from a coordinate in a reference image to a corresponding coordinate in an input image, and matching a local feature quantity extracted from the reference image and a local feature quantity extracted from the input image, an input image different area determination unit for transforming the different area in the reference image on a basis of the geometric transformation information about the input image determined to be in conformity by the matching, and determining a different area in the input image corresponding to the different area in the reference image, an input image different area feature quantity extraction unit for correcting a different area in the input image, and extracting a feature quantity from the corrected different area of the input image, and a feature quantity matching unit for matching a feature quantity extracted by the input image different area feature quantity extraction unit and a feature quantity extracted from the different area in the reference image, and outputting a matching result.

Claims

1. An object identification device comprising:

a memory that stores a set of instructions; and

at least one processor configured to execute the set of instructions to;

calculate geometric transformation information for transformation from coordinates in a reference image to corresponding coordinates in an input image captured electronically, and matches perform matching between a local feature quantity extracted from the reference image and a local feature quantity extracted from the input image;

estimate a different area in the reference image on a basis of density of incorrectly corresponding feature points that is feature points whose feature quantities are determined to incorrectly correspond;

transform the different area in the reference image on a basis of the geometric transformation information calculated for the input image determined to correspond to the reference image by the matching, and determine the transformed area in the reference image to be a different area in the input image corresponding to the different area in the reference image;

correct the different area in the input image, and extracts a feature quantity from the corrected different area of the input image; and

match the feature quantity extracted from the different area in the input image and a feature quantity extracted from the different area in the reference image, and output a matching result indicating an extent to which the input image captured electronically matches the reference image.

2. The object identification device according to claim 1, wherein

the set of instructions is executed by the at least one processor to further correct the different area of the input image into an area enlarged by a preset pixel, and extract the feature quantity from the corrected different area of the input image.

3. The object identification device according to claim 1, wherein

the set of instructions is executed by the at least one processor to further correct the different area of the input image into an area enlarged by a preset ratio on a basis of a reference point in the different area, and extracts the feature quantity from the corrected different area of the input image.

4. The object identification device according to claim 1, wherein

in a case where there is an area where edges are concentrated at an end in the different area of the input image, the set of instructions is executed by the at least one processor to further correct the different area of the input image into an area enlarged in a direction where the edges are concentrated, and extracts the feature quantity from the corrected different area of the input image.

5. The object identification device according to claim 1 wherein the set of instructions is executed by the at least one processor to further:

compare two reference images at a time, and calculates, as the different area in the reference image, an area where there is a difference between compared reference images.

6. The object identification device according to claim 5, wherein the set of instructions is executed by the at least one processor to further:

output geometric transformation information and corresponding feature point information including information indicating a correctly corresponding feature point and an incorrectly corresponding feature point, and

calculate the different area in the reference image on a basis of corresponding feature point information which has been output.

7. The object identification device according claim 1, wherein the set of instructions is executed by the at least one processor to further:

output a first feature quantity which is the feature quantity extracted from the different area in the input image that has been determined, and a second feature quantity which is the feature quantity that has been extracted, and

match the first feature quantity and the feature quantity extracted from the different area in the reference image, and in a case where it is determined that there is no sameness, match the second feature quantity and the feature quantity extracted from the different area in the reference image, and outputs a result of the matching.

8. The object identification device according to claim 7, wherein the set of instructions is executed by the at least one processor to further:

execute, in parallel, matching of the first feature quantity and the feature quantity extracted from the different area in the reference image and matching of the second feature quantity and the feature quantity extracted from the different area in the reference image, and output, as a matching result, a result obtained by integrating the results of the two matchings.

9. An object identification method comprising:

calculating geometric transformation information for transformation from coordinates in a reference image to corresponding coordinates in an input image captured electronically, and matching a local feature quantity extracted from the reference image and a local feature quantity extracted from the input image;

estimating a different area in the reference image on a basis of density of incorrectly corresponding feature points that is feature points whose feature quantities are determined to incorrectly correspond;

transforming the different area in the reference image on a basis of the geometric transformation information calculated for the input image determined to correspond to the reference image by the matching, and determining in the transformed area in the reference image to be a different area in the input image corresponding to the different area in the reference image;

correcting the different area in the input image, and extracting a feature quantity from the corrected different area of the input image; and

matching the feature quantity extracted from the different area of the input image and a feature quantity extracted from the different area in the reference image, and outputting a matching result indicating an extent to which the input image captured electronically matches the reference image.

10. A non-transitory computer readable storage medium storing a computer program causing a computer to perform:

local feature quantity matching processing of calculating geometric transformation information for transformation from coordinates in a reference image to corresponding coordinates in an input image captured electronically, and matching a local feature quantity extracted from the reference image and a local feature quantity extracted from the input image;

estimating processing of estimating a different area in the reference image on a basis of density of incorrectly corresponding feature points that is feature points whose feature quantities are determined to incorrectly correspond;

input image different area determination processing of transforming the different area in the reference image on a basis of the geometric transformation information calculated for the input image determined to correspond to the reference image by the matching, and determining in the transformed area in the reference image to be the different area in the input image corresponding to the transformed different area in the reference image;

input image different area feature quantity extraction processing of correcting the different area in the input image, and extracting a feature quantity from the corrected different area of the input image; and

feature quantity matching processing of matching the feature quantity extracted by the input image different area feature quantity extraction processing and the feature quantity extracted from the different area in the reference image, and outputting a matching result indicating an extent to which the input image captured electronically matches the reference image.