Image recognition device and image recognition method

Patent metadata

Field	Value
Publication number	US-10628709-B2
Application number	US-201815996106-A
Country	US
Kind code	B2
Filing date	Jun 1, 2018
Priority date	Jun 13, 2017
Publication date	Apr 21, 2020
Grant date	Apr 21, 2020

Abstract

An image recognition device includes: a hardware processor that: conducts machine learning, to perform a first process of calculating a plurality of region candidates for a region showing part of an object captured in an image, and a second process of determining a size of each of the region candidates in accordance with the object captured in the image; and determines the region from among the region candidates, using a predetermined criterion.

First claim

What is claimed is:

1. An image recognition device comprising:
a hardware processor that:
conducts machine learning, to perform a first process of calculating a plurality of region candidates for a region showing a part of an object captured in an image, the part of the object being a portion where an outline of a recognition target is not clear, the part of the object that is the recognition target being recognized as a bounding box, and a second process of determining a size of each of the region candidates in accordance with the object captured in the image; and
determines the region from among the region candidates, using a predetermined criterion.

2. The image recognition device according to claim 1 , wherein the hardware processor stores a learning model in advance, and performs the first process and the second process by using the learning model, the learning model being constructed by the hardware processor performing a third process of detecting the region by using a plurality of images in which the region is set and the object is captured, and a fourth process of performing, for each of the images, a process of determining a size of the region in accordance with the object captured in the image, the third process and the fourth process being performed through machine learning.

3. An image recognition device comprising:
a hardware processor that:
conducts machine learning, to perform a first process of calculating a plurality of region candidates for a region showing part of an object captured in an image, and a second process of determining a size of each of the region candidates in accordance with the object captured in the image, and
determines the region from among the region candidates, using a predetermined criterion; and

an inputter that receives an input of a command for setting information indicating that the object is a single object in the hardware processor from an operator of the image recognition device, when the object captured in the image is a single object in a case where the hardware processor is made to perform the first process and the second process,
wherein the hardware processor further:
conducts machine learning, to perform a process of calculating a likelihood that a region candidate is the region, the process being performed for each of the region candidates, and
determines that the region candidate having the highest likelihood among the region candidates is the region.

4. The image recognition device according to claim 1 , further comprising
an inputter that receives an input of a command for setting information indicating that the object is at least two objects in the hardware processor from an operator of the image recognition device, when the object captured in the image is at least two objects in a case where the hardware processor is made to perform the first process and the second process,
wherein the hardware processor conducts machine learning, to perform a process of calculating a plurality of rectangular region candidates for a rectangular region circumscribing the object, for each of the at least two objects captured in the image,
conducts machine learning, to calculate a classification probability indicating which of the rectangular region candidates a region candidate belongs to, for each of the region candidates,
determines the rectangular region circumscribing the object from among the rectangular region candidates, for each of the at least two objects,
conducts machine learning, to perform a process of calculating a likelihood that a region candidate is the region, for each of the region candidates, and
determines the region belonging to the rectangular region from among the region candidates, for each of the two or more rectangular regions, in accordance with the classification probabilities of the region candidates belonging to the rectangular region candidate determined to be the rectangular region, and the likelihoods of the respective region candidates.

5. An image recognition method comprising:
conducting machine learning, to perform a first process of calculating a plurality of region candidates for a region showing a part of an object captured in an image, the part of the object being a portion where an outline of a recognition target is not clear, the part of the object that is the recognition target being recognized as a bounding box, and a second process of determining a size of each of the region candidates in accordance with the object captured in the image; and
determining the region from among the region candidates, using a predetermined criterion.

6. A non-transitory computer readable medium storing instructions to cause a processor-controlled apparatus to perform the method of claim 5 .

Assignees

• Konica Minolta Inc

Inventors

• Sekii Taiki

Classifications

• G06T2210/12

  Bounding box · CPC title

• G06T7/11

  Region-based segmentation · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T7/187

  involving region growing; involving region merging; involving connected component labelling · CPC title

• G06K9/00375

  Physics · mapped topic