Currency classification device and currency classification method

Patent metadata

Field	Value
Publication number	US-11017273-B2
Application number	US-201716338815-A
Country	US
Kind code	B2
Filing date	Sep 27, 2017
Priority date	Oct 7, 2016
Publication date	May 25, 2021
Grant date	May 25, 2021

Abstract

A currency classification device that classifies currency types using currency images includes a feature value calculator, a storage, and an output unit. The feature value calculator calculates feature values for every currency type that is a candidate for classification from an image area common in images of every currency type. The storage stores the feature values calculated by the feature value calculator from learning images, which are currency models, as templates for every currency type. The output unit outputs the currency type corresponding to the template having a highest value of similarity with the feature value calculated by the feature value calculator an input image, which is a currency image subject to classification, in the templates stored in the storage as a classification result.

First claim

The invention claimed is:

1. A currency classification device that classifies currency types using currency images, the device comprising:
a circuitry that calculates feature values for every currency type that is a candidate for classification from an image area, which is a part of each currency image, the image area being defined by a radial direction from a reference point and an angular direction about the reference point of the currency image, wherein a location of the image area subjected to calculation for the feature value is common in images of every currency type; and
a storage that stores the feature values calculated by the circuitry from learning images, which are currency models, as templates for every currency type,
wherein
the circuitry outputs the currency type corresponding to the template having a highest value of similarity with the feature value calculated by the circuitry from an input image, which is a currency image subject to classification, in the templates stored in the storage as a classification result,
the circuitry calculates an autocorrelation function curve that has a coordinate axis extending in the angular direction about the reference point of a currency image as the feature value for each position in the radial direction from the reference point of the currency image, and
the similarity is calculated based on synchronization features of the autocorrelation function curve calculated by the circuitry from the input image and the autocorrelation function curve calculated by the circuitry from the learning image for each currency type.

2. The currency classification device according to claim 1 , wherein the circuitry outputs the currency type corresponding to the template having a similarity that is greater than or equal to a predetermined threshold value including the currency type corresponding to the template having the highest value of similarity as the classification result.

3. The currency classification device according to claim 1 , wherein the circuitry outputs classification results of the currency types corresponding to the templates in order of similarity from higher ones including the currency type corresponding to the template having the highest value of similarity.

4. The currency classification device according to claim 1 , wherein
the circuitry calculates the feature values based on different calculation conditions, and
the similarity is calculated using all of the feature values calculated by the circuitry from the input image that is the currency image subject to classification.

5. The currency classification device according to claim 1 , wherein
the circuitry calculates the feature values based on different calculation conditions,
the template is calculated as an average value or a median value of the feature values calculated from learning images corresponding to the same currency type, and
the similarity is calculated by comparing dispersion degrees of the feature values used to calculate the template of each currency type for each of the calculation conditions and giving priority to the feature values calculated based on the calculation conditions having relatively small dispersion degrees.

6. The currency classification device according to claim 1 , wherein the circuitry calculates the feature values based on an edge gradient component extracted from a currency image.

7. The currency classification device according to claim 6 , wherein the circuitry extracts edge gradient components having different edge directions and calculates the feature values based on ratios of the extracted edge gradient components.

8. The currency classification device according to claim 6 , wherein the circuitry calculates the feature values based on an edge strength component extracted from the currency image.

9. The currency classification device according to claim 1 , wherein the circuitry adds image data of multiple positions in the radial direction from the reference point of the currency image for each position in the angular direction about the reference point of the currency image and calculates the autocorrelation function curve for each position in the radial direction from the reference point of the currency image based on the image data obtained through the addition.

10. The currency classification device according to claim 1 , wherein the circuitry calculates the feature values without changing the location of the image area of a subject currency image.

11. The currency classification device according to claim 1 , wherein
the image area that is common in images is a single image area.

12. The currency classification device according to claim 1 , wherein
the image area that is common in images of every currency type is set to a predetermined region in a radial direction of the learning images and the currency image subject to the classification.

13. A method for classifying types of currency using currency images, the method comprising:
calculating feature values for every currency type that is a candidate for classification from an image area, which is a part of each currency image, the image area being defined by a radial direction from a reference point and an angular direction about the reference point of the currency image, wherein a location of the image area subjected to calculation for the feature value is common in images of every currency type;
reading the feature values calculated from learning images, which are currency models, as templates for every currency type;
outputting the currency type corresponding to the template having a highest value of similarity with the feature values calculated from an input image, which is a currency image subject to classification, in the read templates as a classification result; and
calculating an autocorrelation function curve that has a coordinate axis extending in the angular direction about the reference point of a currency image as the feature value for each position in the radial direction from the reference point of the currency image,
wherein the similarity is calculated based on synchronization features of the autocorrelation function curve from the input image and the autocorrelation function curve from the learning image for each currency type.

Assignees

• Glory Kogyo Kk

Inventors

• Morishita Ruka
• Yonezawa Toru
• Tada Kazuya
• Manabe Hideaki

Classifications

• G07D5/005 Primary

  Testing the surface pattern, e.g. relief · CPC title

• G06F18/24

  Classification techniques · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• G07D7/2008

  using pre-processing, e.g. de-blurring, averaging, normalisation or rotation · CPC title

• G07D7/206

  Matching template patterns · CPC title