Search

JP-7856294-B2 - Purchase data output system

JP7856294B2JP 7856294 B2JP7856294 B2JP 7856294B2JP-7856294-B2

Inventors

  • 安藤 秀之

Assignees

  • キャッシュビーデータ株式会社

Dates

Publication Date
20260511
Application Date
20220412

Claims (4)

  1. A purchase data output system that outputs product purchase data, The first machine learning unit analyzes the relationship between a receipt image and the product information and other information listed in the receipt image using machine learning. A second machine learning unit analyzes the relationship between the product information listed in the aforementioned receipt image and the official name information of the product registered in a predetermined database using machine learning. An image processing unit that performs edge detection and plane transformation on a receipt image acquired from a user terminal, and then applies either distortion correction or noise reduction to the image processing, Based on the relationships analyzed by the first machine learning unit, the first estimation unit takes the processed receipt image as input and estimates and outputs product information and non-product information described in the receipt image. Based on the relationships analyzed by the second machine learning unit, the second estimation unit takes the product information output from the first estimation unit as input and estimates and outputs the official name information of the product. Equipped with, Information other than the aforementioned product includes any of the following: the company name or brand name of the store selling the product, the name of the store selling the product, the store's telephone number or address, the date or time of purchase of the product, the quantity or unit price or subtotal or total or tax of the product, or the payment method used when purchasing the product. The aforementioned purchase data output system is A purchase data creation unit creates purchase data for the product by adding information other than the product output from the first estimation unit to the information of the official name of the product output from the second estimation unit, Equipped with , The system includes a first information acquisition unit that obtains information on the correct store name, correct store telephone number, or correct store address of a store that sells the product by performing an information search based on at least one piece of information output from the first estimation unit, which is the store name, store telephone number, or store address of the store that sells the product. The purchase data creation unit adds information about the correct store name, correct store telephone number, or correct store address that sells the product to the information about the official name of the product output from the second estimation unit, thereby creating purchase data for the product .
  2. A purchase data output system that outputs product purchase data, The first machine learning unit analyzes the relationship between a receipt image and the product information and other information listed in the receipt image using machine learning. A second machine learning unit analyzes the relationship between the product information listed in the receipt image and the official name information of the product registered in a predetermined database using machine learning. An image processing unit that performs edge detection and plane transformation on a receipt image acquired from a user terminal, and then applies either distortion correction or noise reduction to the image processing, Based on the relationships analyzed by the first machine learning unit, the first estimation unit takes the processed receipt image as input and estimates and outputs product information and non-product information described in the receipt image. Based on the relationships analyzed by the second machine learning unit, the second estimation unit takes the product information output from the first estimation unit as input and estimates and outputs the official name information of the product. Equipped with, Information other than the aforementioned product includes any of the following: the company name or brand name of the store selling the product, the name of the store selling the product, the store's telephone number or address, the date or time of purchase of the product, the quantity or unit price or subtotal or total or tax of the product, or the payment method used when purchasing the product. The aforementioned purchase data output system is A purchase data creation unit creates purchase data for the product by adding information other than the product output from the first estimation unit to the information of the official name of the product output from the second estimation unit, Equipped with , The system includes a second information acquisition unit that, based on the information of the official name of the product output from the second estimation unit, obtains the manufacturer name or brand name of the manufacturer of the product, or the category or JAN code information of the product, by referring to the predetermined database. The purchase data creation unit is a purchase data output system that creates purchase data for a product by adding the manufacturer's name or brand name, or the product's category or JAN code information, to the information of the official name of the product output from the second estimation unit.
  3. A method performed by a purchase data output system that outputs product purchase data, The aforementioned method, The first machine learning step involves analyzing the relationship between a receipt image and the product information and other information listed in the receipt image using machine learning. A second machine learning step involves analyzing the relationship between the product information listed on the receipt image and the official name information of the product registered in a predetermined database, using machine learning. Image processing steps include: performing edge detection and plane transformation on a receipt image acquired from a user terminal, followed by image processing of either distortion correction or noise reduction; Based on the relationships analyzed in the first machine learning step, the first estimation step takes the processed receipt image as input and estimates and outputs product information and non-product information described in the receipt image. Based on the relationships analyzed in the second machine learning step, the second estimation step takes the product information output in the first estimation step as input and estimates and outputs the official name information of the product. Equipped with, Information other than the aforementioned product includes any of the following: the company name or brand name of the store selling the product, the name of the store selling the product, the store's telephone number or address, the date or time of purchase of the product, the quantity or unit price or subtotal or total or tax of the product, or the payment method used when purchasing the product. The aforementioned method, A purchase data creation step involves creating purchase data for the product by adding information other than the product output in the first estimation step to the information of the official name of the product output in the second estimation step, Includes, The above method further, The process includes a first information acquisition step in which information is obtained by performing an information search based on at least one of the following pieces of information output in the first estimation step: the name of the store that sells the product, the store's telephone number, or the store's address, thereby obtaining the correct name of the store that sells the product, the correct telephone number, or the correct address of the store that sells the product. The purchase data creation step involves adding information about the correct store name, correct store telephone number, or correct store address that sells the product to the information about the official name of the product output in the second estimation step, thereby creating purchase data for the product .
  4. A method performed by a purchase data output system that outputs product purchase data, The aforementioned method, The first machine learning step involves analyzing the relationship between a receipt image and the product information and other information listed in the receipt image using machine learning. A second machine learning step involves analyzing the relationship between the product information listed on the receipt image and the official name information of the product registered in a predetermined database, using machine learning. Image processing steps include: performing edge detection and plane transformation on a receipt image acquired from a user terminal, followed by image processing of either distortion correction or noise reduction; Based on the relationships analyzed in the first machine learning step, the first estimation step takes the processed receipt image as input and estimates and outputs product information and non-product information described in the receipt image. Based on the relationships analyzed in the second machine learning step, the second estimation step takes the product information output in the first estimation step as input and estimates and outputs the official name information of the product. Equipped with, Information other than the aforementioned product includes any of the following: the company name or brand name of the store selling the product, the name of the store selling the product, the store's telephone number or address, the date or time of purchase of the product, the quantity or unit price or subtotal or total or tax of the product, or the payment method used when purchasing the product. The aforementioned method, A purchase data creation step involves creating purchase data for the product by adding information other than the product output in the first estimation step to the information of the official name of the product output in the second estimation step, Includes, The above method further, Based on the information of the official name of the product output in the second estimation step, the process includes a second information acquisition step in which the manufacturer name or brand name of the manufacturer of the product, or the category or JAN code information of the product is obtained by referring to the predetermined database. The purchase data creation step involves adding the manufacturer's name or brand name, or the product's category or JAN code information, to the official name information of the product output in the second estimation step to create purchase data for the product .

Description

This invention relates to a purchase data output system that outputs product purchase data. Conventionally, technologies have been proposed to digitize information printed on receipts. For example, a system for analyzing information printed on receipts has been proposed (see Patent Document 1). In conventional systems, the information contained in the information printed on the receipt (text data) is broken down into words, and based on the positional relationships of the words, words or groups of words are categorized. Japanese Patent Publication No. 2017-120521 This is a block diagram showing the configuration of the purchase data output system in this embodiment.This block diagram shows an example of the configuration of the image processing unit in this embodiment.This diagram shows an example of product information and other information listed on a receipt.This figure shows an example of product data registered in the database.This is a flowchart illustrating the operation of the purchase data output system in this embodiment. The following describes an embodiment of the purchase data output system of the present invention with reference to the drawings. This embodiment illustrates a purchase data output system used in systems that digitize receipt information, etc. The configuration of the purchase data output system according to an embodiment of the present invention will be described with reference to the drawings. Figure 1 is a block diagram showing the configuration of the purchase data output system according to this embodiment. As shown in Figure 1, the purchase data output system 1 is connected to a user terminal 2 used by a general user purchasing goods, a database 3 in which product data (described later) related to the goods is registered, and a network N such as the Internet. The purchase data output system 1 is composed of, for example, a server device. Furthermore, the purchase data output system 1 has functions for outputting product purchase data, and these functions can be realized by programs installed in the memory of the purchase data output system 1 (server device). As shown in Figure 1, the purchase data output system 1 comprises an input/output unit 4, a storage unit 5, a first system 6, a second system 7, and a third system 8. The input/output unit 4 is an input/output interface for inputting and outputting various types of data for outputting product purchase data. While this example describes an integrated input and output interface, the input and output interfaces may be configured separately. The storage unit 5 is composed of memory and stores various types of data for outputting product purchase data. Furthermore, as shown in Figure 1, the first system 6 comprises a first machine learning unit 9, an image processing unit 10, and a first estimation unit 11. The second system 7 comprises a second machine learning unit 12 and a second estimation unit 13. The third system 8 comprises a first information acquisition unit 14, a second information acquisition unit 15, and a purchase data creation unit 16. Also, as shown in Figure 2, the image processing unit 10 comprises an edge detection unit 17, a plane transformation unit 18, a binarization unit 19, a smoothing unit 20, and a noise reduction unit 21. The first machine learning unit 9 analyzes the relationship between a receipt image and the product information and other information listed on that receipt image, using machine learning. This machine learning can employ any method, such as deep learning using neural networks. For example, in a neural network, a receipt image is input to the input layer, and information about the products and other information listed on the receipt image is output from the output layer. Then, using supervised learning with analysis data (training data) that links the input data (receipt image) to the output data (product information and other information), the weighting coefficients between neurons in the neural network are optimized. The input/output unit 4 receives the receipt image to be judged (the receipt image acquired from the user terminal 2). The image processing unit 10 performs image processing (preprocessing) on the input receipt image (the receipt image to be judged) to make it easier to estimate product information and other information from the receipt image. First, the edge detection unit 17 performs edge detection on the input receipt image. For example, edge detection extracts an image of a rectangular image area from the input receipt image that includes the image region where the receipt is visible. Known techniques can be used for edge detection. Next, in the image processing unit 10, the plane transformation unit 18 performs a process (plane transformation process) to convert the image of the rectangular image area extracted by edge detection (an image of the receipt viewed from the front, an image with perspective) into a plane image of the receipt viewed from the front (an i