Image feature extraction method, image feature extraction apparatus, and recording medium storing program therefor

Patent metadata

Field	Value
Publication number	US-9600737-B2
Application number	US-201514945920-A
Country	US
Kind code	B2
Filing date	Nov 19, 2015
Priority date	Nov 28, 2014
Publication date	Mar 21, 2017
Grant date	Mar 21, 2017

Abstract

A feature extraction method includes acquiring an image, detecting a vanishing point in the image, setting a plurality of ellipses and a plurality of half-lines with respect to the detected vanishing point to segment the image into a plurality of regions, and extracting a feature of each segmented region. With the disclosed method, a global feature can be appropriately extracted from an image including a vanishing point or other relevant features.

First claim

What is claimed is:

1. A feature extraction method comprising:
acquiring an image;
detecting a vanishing point in the acquired image;
setting m ellipses and n half-lines centered on the detected vanishing point to segment the image into a plurality of regions;
extracting a feature of a region segmented by the set ellipses and half-lines; and
classifying the image into a plurality of groups by extracting features of the segmented regions and clustering the plurality of extracted features into the plurality of groups,
wherein, in a case where m=0, n is an integer satisfying n≧2, and in a case where m is an integer satisfying m≧1, n is an integer satisfying n≧0.

2. The feature extraction method according to claim 1 , wherein in the setting, the plurality of ellipses is set so that x-direction radii and y-direction radii are arranged at regular intervals.

3. The feature extraction method according to claim 1 , wherein in the setting, the plurality of half-lines is set so that inclination angles between adjacent half-lines are formed at regular intervals.

4. The feature extraction method according to claim 1 , wherein in the extracting, any one of color information, geometric feature, color histogram, and binary pattern feature is extracted as a feature of the segmented region.

5. The feature extraction method according to claim 1 , wherein in the classifying, a label of a small region of the image is identified based on the features of the segmented regions of the image divided into the plurality of groups by use of information about a correct answer label and a feature of the small region of the image which is held with respect to each of the plurality of groups.

6. The feature extraction method according to claim 1 , further comprising learning, by use of an image with a correct answer label, information about the correct answer label and a feature of a small region of the image.

7. The feature extraction method according to claim 1 , further comprising:
dividing the acquired image into a plurality of blocks and extracting a feature from the plurality of blocks in a case where the vanishing point is not detected from the acquired image.

8. A feature extraction apparatus comprising:
at least one processor connected to a memory and configured to execute instructions that, when executed, cause the feature extraction apparatus to:
acquire an image;
detect a vanishing point in the image acquired by an acquisition unit;
set m ellipses and n half-lines centered on the vanishing point to segment the image into a plurality of regions;
extract a feature of a region segmented by the ellipses and the half-lines set by a setting unit; and
classify the image into a plurality of groups by extracting features of the segmented regions and clustering the plurality of extracted features into the plurality of groups,
wherein, in a case where m=0, n is an integer satisfying n≧2, and in a case where m is an integer satisfying m≧1, n is an integer satisfying n≧0.

9. A non-transitory computer-readable recording medium that stores a program for causing a computer to execute a feature extraction process comprising:
acquiring an image;
detecting a vanishing point in the acquired image;
setting m ellipses and n half-lines centered on the vanishing point to segment the image into a plurality of regions;
extracting a feature of a region segmented by the set ellipses and the set half-lines; and
classifying the image into a plurality of groups by extracting features of the segmented regions and clustering the plurality of extracted features into the plurality of groups,
wherein, in a case where m=0, n is an integer satisfying n≧2, and in a case where m is an integer satisfying m≧1, n is an integer satisfying n≧0.

Assignees

• Canon Kk

Inventors

• Iwasaki Yasushi

Classifications

• G06T7/10 Primary

  Segmentation; Edge detection (motion-based segmentation G06T7/215) · CPC title

• G06F18/24

  Classification techniques · CPC title

• G06K9/46 Primary

  Physics · mapped topic

• G06T7/0079

  Physics · mapped topic

• G06K9/6267

  Physics · mapped topic