Learning image automatic sorting device, learning image automatic sorting method, and learning image automatic sorting program

Patent metadata

Field	Value
Publication number	US-10679358-B2
Application number	US-201816015834-A
Country	US
Kind code	B2
Filing date	Jun 22, 2018
Priority date	Dec 22, 2015
Publication date	Jun 9, 2020
Grant date	Jun 9, 2020

Abstract

A learning image automatic sorting device for sorting learning images to be stored while identifying and eliminating repeated images by discriminating the geometric deformation, changes in grayscale, the difference in noise level, and the like.

First claim

The invention claimed is:

1. A learning image automatic sorting device that automatically sorts learning images used for image recognition processing from a plurality of captured images captured in time series, comprising:
movement detection means for detecting a movement amount of a subject in an image of interest using the image of interest which is a sorting determination target among the plurality of captured images and an image captured temporally earlier than the image of interest;
a decision unit that decides whether or not to select the image of interest as a learning candidate image based on the movement amount;
a derivative image generation unit that generates one or a plurality of derivative images using at least one of the image of interest selected as the learning candidate image by the decision unit and the learning image;
a similarity calculation unit that calculates a similarity between the learning candidate image and the learning image and a similarity between the one or the plurality of derivative images and the learning image, or a similarity between the one or the plurality of derivative images and the learning candidate image;
a sorting unit that sorts the image of interest selected as the learning candidate image, as a new learning image based on the similarity calculated by the similarity calculation unit and the statistically determined condition; and
a storage unit that stores the learning image.

2. The learning image automatic sorting device according to claim 1 ,
wherein the sorting unit compares the similarity with a threshold value, and in a case where a predetermined ratio of the similarity to the value of the similarity calculated by the similarity calculation unit is less than the threshold value, the sorting unit sorts the image of interest selected as the learning candidate image as a new learning image.

3. The learning image automatic sorting device according to claim 1 ,
wherein the derivative image generation unit generates the derivative image by performing at least one of image conversion processing selected from a group of a geometric conversion, a grayscale conversion, a noise increase/decrease, a saturation enhancement, an edge enhancement, and a color conversion on the learning candidate image.

4. The learning image automatic sorting device according to claim 1 , further comprising:
a detection unit that detects shake information or blur information on the image of interest,
wherein the decision unit excludes the image of interest from the learning candidate image in a case where a shake or a blur is detected from the image of interest using shake information or blur information.

5. The learning image automatic sorting device according to claim 1 , further comprising:
a setting unit that performs setting of a target image for generating the derivative image among the learning candidate image and the learning image.

6. The learning image automatic sorting device according to claim 1 ,
wherein the derivative image generation unit generates the one or the plurality of derivative images using the image of interests selected as the learning candidate image by the decision unit.

7. A learning image automatic sorting method for automatically sorting a learning image used for image recognition processing from a plurality of captured images captured in time series, comprising:
detecting a movement amount of a subject in an image of interest using the image of interest which is the sorting determination target among the plurality of captured images and an image captured temporally earlier than the image of interest;
deciding whether or not to select the image of interest as a learning candidate image based on the movement amount;
generating one or a plurality of derivative images using at least one of the image of interest selected as the learning candidate image by the decision unit and the learning image;
calculating a similarity between the learning candidate image and the learning image and a similarity between the one or the plurality of derivative images and the learning image, or a similarity between the one or the plurality of derivative images and the learning candidate image;
sorting the image of interest selected as the learning candidate image, as a new learning image based on the similarity calculated by the similarity calculation unit and the statistically determined condition; and
storing the image of interest sorted as the selected learning candidate image as the learning image.

8. A non-transitory computer-readable storage medium storing a learning image automatic sorting program for automatically sorting a learning image used for image recognition processing from a plurality of captured images captured in time series, wherein the program causes a computer to execute:
a detection procedure for detecting the movement amount of a subject in an image of interest using the image of interest which is the sorting determination target among the plurality of captured images and an image captured temporally earlier than the image of interest;
a decision procedure for deciding whether or not to select the image of interest as a learning candidate image based on the movement amount;
a derivative image generation procedure for generating one or a plurality of derivative images using at least one of the image of interest selected as the learning candidate image by the decision unit and the learning image;
a similarity calculation procedure for calculating a similarity between the learning candidate image and the learning image and a similarity between the one or the plurality of derivative images and the learning image, or a similarity between the one or the plurality of derivative images and the learning candidate image;
a sorting procedure for sorting the image of interest selected as the learning candidate image, as a new learning image based on the similarity and statistically determined conditions; and
a storing procedure for storing the image of interest selected as the selected learning candidate image as the learning image.

Assignees

• Olympus Corp

Inventors

• On Seigo

Classifications

• G06K9/6255

  Physics · mapped topic

• G06T7/254

  involving subtraction of images · CPC title

• G06T2207/30096

  Tumor; Lesion · CPC title

• G06T7/74

  involving reference images or patches · CPC title

• G06T2207/20081

  Training; Learning · CPC title