Feature extraction method, object classification method, object identification method, feature extraction device, object classification device, object identification device, feature extraction/object classification/object identification program, and recording medium on which the program is recorded

Patent metadata

Field	Value
Publication number	US-9230313-B2
Application number	US-201214365811-A
Country	US
Kind code	B2
Filing date	Jul 18, 2012
Priority date	Dec 16, 2011
Publication date	Jan 5, 2016
Grant date	Jan 5, 2016

Abstract

Provided is a feature extraction that extracts a feature that represents a characteristic of a subject, the feature being extracted from an image that has imaged the subject, the feature being extracted without relation to the shape of the subject. The feature extraction extracts the feature from the image of the subject, the subject having been imaged by an imaging means.

First claim

The invention claimed is:

1. A feature extraction method for extracting a feature description from an image of a subject, the feature extraction method comprising the steps of:
capturing the image with an image capturing device;
electrically creating a filter bank from said image;
electrically creating a maximum brightness image from said filter bank;
electrically setting a circular image region of said maximum brightness image, and setting a center Cc and a radius Rc of the circular image region;
electrically projecting a pixel in the maximum brightness image in a three-dimensional space having axes representing (a) a ratio between a distance L D between a pixel position (x, y) of the pixel and said center Cc, and said radius Rc, (b) a brightness value F D (x, y) of the pixel, and (c) the total of the difference between the brightness value F D of the pixel and the brightness value of a nearby pixel; and
electrically creating a frequency histogram from the pixels projected in said three-dimensional space.

2. The feature extraction method according to claim 1 , said subject being a soft item.

3. The feature extraction method according to claim 1 , said filter bank being created using a Gabor filter.

4. An object classification method comprising:
electrically classifying the subject using the frequency histogram extracted using the feature extraction method according to claim 1 .

5. An object identification method comprising;
electrically comparing the frequency histogram extracted using the feature extraction method according to claim 1 to a frequency histogram of a known subject; and
identifying a type of the subject.

6. The object identification method according to claim 5 , wherein the comparing the extracted frequency histogram with a frequency histogram of a plurality of the known subject types, whereby the type of the subject is identified.

7. A computer program stored in a non-transitory computer readable medium, the computer program configuring a feature extraction device to execute the feature extraction method according to claim 1 .

8. A non-transitory computer-readable medium that, when executed with a feature extraction device, causes the feature extraction device to execute the feature extraction method according to claim 1 .

9. A feature extraction method for extracting a feature description from an image of a subject, the feature extraction method comprising:
capturing the image with an image capturing device;
electrically creating a filter bank from said image;
electrically creating a maximum brightness image from said filter bank;
electrically setting a circular image region of said maximum brightness image, and setting a center Cc and a radius Rc of the circular image region;
electrically projecting a pixel in the maximum brightness image in a three-dimensional space having axes representing (d) a ratio between the distance L o between the position (x, y) of the pixel and said center Cc, and said radius Rc, (e) a value E o in which whether the pixel is positioned on an upper side or a lower side of a folded overlap is evaluated by a continuous value, and (f) a direction component of the folded overlap portion in which the pixel is present; and
electrically creating a frequency histogram from the pixel projected in said three -dimensional space.

10. A feature extraction device for extracting a feature description from an image of a subject captured by image-capturing means, the feature extraction device comprising:
filter bank creation means for creating a filter bank from said image;
filtering result synthesizing means for creating a maximum brightness image from said filter bank;
maximum brightness image center and radius setting means for setting a circular image region of said maximum brightness image, and setting a center Cc and a radius Rc of the circular image region;
three-dimensional projection means for projecting a pixel in the maximum brightness image in a three-dimensional space having axes representing (a) a ratio between the distance L D between the pixel position (x, y) of the pixel and said center Cc, and said radius Rc, (b) the brightness value F D (x, y) of the pixel, and (c) the total of the difference between the brightness value F D of the pixel and the brightness value of a nearby pixel; and
frequency histogram creation means for creating a frequency histogram from the pixels projected in said three-dimensional space.

11. The feature extraction device according to claim 10 , wherein said subject being a soft item.

12. The feature extraction device according to claim 10 , wherein said filter bank being created using a Gabor filter.

13. An object classification device comprising image classification means for classifying a subject using the frequency histogram extracted using the feature extraction device according to claim 10 .

14. An object identification device comprising:
an identification database storing a frequency histogram of a known subject, and
identification means for comparing, with the frequency histogram of the known subject stored in said identification database, and identifying, the frequency histogram extracted by the feature extraction device according to claim 10 .

15. The object identification device according to claim 14 , wherein:
the frequency histograms of the known subject stored in said identification database are frequency histograms of a plurality of types of subjects, and
said identification means compares said extracted frequency histogram with the plurality of types of frequency histograms stored in said identification database, and thereby identifies the type of the subject.

Assignees

• Yamazaki Kimitoshi
• Inaba Masayuki
• Japan Science & Tech Agency

Inventors

• Yamazaki Kimitoshi
• Inaba Masayuki

Classifications

• G06T5/40 Primary

  using histogram techniques · CPC title

• G06V10/70

  using pattern recognition or machine learning (optical pattern recognition or electronic computations therefor G06V10/88) · CPC title

• G06V10/60

  relating to illumination properties, e.g. using a reflectance or lighting model · CPC title

• G06V10/50

  by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis · CPC title

• G06T2207/30181

  Earth observation · CPC title