# US 2023/0053838 A1 · Image recognition apparatus, image recognition method, and recording medium

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2023/0053838 A1
- Document ID: US20230053838A1
- Status: Published application
- Kind: A1
- Application number: 17/796,702
- Filed: 2020-02-18
- Published: 2023-02-23
- Earliest claimed: 2020-02-18
- Assignees: NEC Corporation
- Applicants: NEC Corporation
- Inventors: Shigeaki NAMIKI; Shoji YACHDA; Takashi SHIBATA
- Web page: https://patentgrep.com/patent/US20230053838A1

## CPC

- G06T 7/0004: {Industrial image inspection}
- G06T 7/11: Region-based segmentation
- G06T 7/194: involving foreground-background segmentation
- G06T 2207/10016: Video; Image sequence
- G06T 2207/10152: Varying illumination
- G06T 2207/20021: Dividing image into blocks, subimages or windows
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

The image recognition apparatus includes an image selection unit and a recognition unit. The image selection unit selects a feature image representing a feature portion of an object from among captured images of a time series in which the object is photographed. For example, the feature image corresponds to an image showing an abnormal portion. The recognition unit performs a recognition process of the object using the feature image. By the recognition process, an abnormality of the object is detected.

## Claims

### Claim 1

An image recognition apparatus comprising:
a memory storing instructions; and
one or more processors configured to execute the instructions to:
select a feature image representing a feature portion of an object from among captured images of a time series in which the object is photographed; and
perform a recognition process of the object using the feature image.

### Claim 2 (depends on 1)

The image recognition apparatus according to claim 1, wherein the processor divides each of the captured images into a plurality of cells, and selects the feature image from among the captured images of the time series based on a change of a statistical quantity for each of the plurality of cells.

### Claim 3 (depends on 2)

The image recognition apparatus according to claim 2, wherein the processor selects successively captured images as feature images from a captured image at which the change of the statistical quantity for each of the plurality of cells has begun to a captured image in which the change has ends.

### Claim 4 (depends on 2)

The image recognition apparatus according to claim 2, wherein the cells are either of cells of a predetermined size in which the captured image is divided, super pixel cells, and pixels forming the captured image.

### Claim 5 (depends on 1)

The image recognition apparatus according to claim 1, wherein the processor uses a neural network, and selects the feature image using a trained model so as to select the feature image from among the captured images of the time series.

### Claim 6 (depends on 5)

The image recognition apparatus according to claim 5, wherein the processor extracts feature amounts from the captured images of the time series, generates a vector representing non-redundancy degrees among the captured images of the time series based on the feature amounts, and selects the feature image from among the captured images of the time series using the vector.

### Claim 7 (depends on 6)

The image recognition apparatus according to claim 6, wherein the processor divides each of the captured images into a plurality of cells, and selects the feature image from among the captured images of the time series based on the non-redundancy degrees of respective cells for each of the captured images.

### Claim 8 (depends on 7)

The image recognition apparatus according to claim 7, wherein
the processor outputs attention information of a cell that is a basis for selecting the feature image; image, to the recognition unit; and
the processor recognizes the feature portion of the object using the attention information.

### Claim 9 (depends on 5)

The image recognition apparatus to claim 5, wherein the processor performs an image selection and the recognition process which are formed by a single neural network.

### Claim 10 (depends on 1)

The image recognition apparatus to claim 1, wherein
the feature portion is a portion that indicates an abnormality existing on the object; and
the processor performs a class classification concerning the abnormality of the object, or a detection of the abnormality existing on the object.

### Claim 11

An image recognition method comprising:
selecting a feature image representing a feature portion of an object from among captured images of a time series in which the object is photographed; and
performing a recognition process of the object using the feature image.

### Claim 12

A non-transitory computer-readable recording medium storing a program, the program causing a computer to perform a process comprising:
selecting a feature image representing a feature portion of an object from among captured images of a time series in which the object is photographed; and
performing a recognition process of the object using the feature image.
