Granted Patent B2

US 12,394,180 · App. 17/814,030 · Granted Aug 19, 2025

Inventors: Takuya Miyamoto (Osaka, JP); Kanako Morimoto (Osaka, JP); Rui Hamabe (Osaka, JP); Shiro Kaneko (Osaka, JP); Naomichi Higashiyama (Osaka, JP)

Assignee: KYOCERA DOCUMENT SOLUTIONS INC.

G06V10/7715G06V10/26

Abstract

Claims (36)

1. A computer-implemented image recognition method, comprising:

feature-amount extraction of generating a base feature-map group constituted by a plurality of base feature maps from an input image, and applying a plurality of types of statistic calculations to the base feature maps in the base feature-map group thereby generating a plurality of statistic maps;

inference of deriving inference results of segmentation using an inference unit for inference inputs based on the plurality of statistic maps; and

an image recognition of generating an image recognition result of the input image based on the inference result of the segmentation,

wherein each of the plurality of types of statistic calculations is processing of calculating a statistic with a specific window size and a specific calculation formula, and the plurality of types of statistic calculations are different from each other in at least either one of the window size and the calculation formula, and

wherein in the feature-amount extraction, the plurality of statistic maps are generated based on the plurality of types of statistic calculations consisting of the window size and the calculation formula that are different from each other in at least one aspect.

2. The image recognition method according to claim 1 , wherein

the base feature map has two-dimensional data indicating positions, sizes, and directions of a plurality of specific shapes, and

the plurality of inference inputs are one or a plurality of statistic maps classified by the size.

3. The image recognition method according to claim 1 , wherein

the inference unit is a machine-learned inference unit.

4. The image recognition method according to claim 1 , wherein

the inference unit generates the inference results by clustering without using machine learning.

5. The image recognition method according to claim 1 , further comprising:

integration, wherein

in the inference, a plurality of inference results are derived for the plurality of inference inputs based on the plurality of statistic maps by using each of the plurality of the inference units, and in the integration, a final inference result is derived by integrating the plurality of inference results in a specific method.

6. The image recognition method according to claim 5 , wherein,

each of the plurality of inference inputs has some or all of the plurality of statistic maps; and

each inference input in the plurality of inference inputs has a statistic map partially or wholly different from the statistic maps of the other inference inputs in the plurality of inference inputs.

7. The image recognition method according to claim 5 , further comprising:

inference input generation of generating the plurality of inference inputs from the statistic map group, wherein

each of the plurality of base feature maps is extracted from the input image in the plural specific processing; and

the inference input has one or a plurality of statistic maps selected from the plurality of statistic maps corresponding to the plural specific processing.

8. An image recognition apparatus, comprising:

a feature-amount extraction unit which generates a base feature-map group constituted by a plurality of base feature maps from an input image, and applies a plurality of types of statistic calculations to the base feature maps in the base feature-map group thereby generating a plurality of statistic maps;

an inference unit which derives inference results of segmentation by an inference unit for inference inputs based on the plurality of statistic maps; and

an image recognition unit which generates an image recognition result of the input image based on the inference result of the segmentation,

wherein each of the plurality of types of statistic calculations is processing of calculating a statistic with a specific window size and a specific calculation formula, and the plurality of types of statistic calculations are different from each other in at least either one of the window size and the calculation formula, and

wherein the feature amount extraction unit generates the plurality of statistic maps based on the plurality of types of statistic calculations consisting of the window size and the calculation formula that are different from each other in at least one aspect.

9. A computer-readable non-transitory recording medium storing an image recognition program, wherein

the image recognition program causes a computer to function as:

a feature-amount extraction unit which generates a base feature-map group constituted by a plurality of base feature maps from an input image, and applies a plurality of types of statistic calculations to the base feature maps in the base feature-map group thereby generating a plurality of statistic maps;

an inference unit which derives inference results of segmentation by the inference unit for the inference inputs based on the plurality of statistic maps; and

an image recognition unit which generates an image recognition result of the input image based on the inference result of the segmentation,

wherein each of the plurality of types of statistic calculations is processing of calculating a statistic with a specific window size and a specific calculation formula; and the plurality of types of statistic calculations are different from each other in at least either one of the window size and the calculation formula, and

wherein the feature amount extraction unit generates the plurality of statistic maps based on the plurality of types of statistic calculations consisting of the window size and the calculation formula that are different from each other in at least one aspect.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (16)