Granted Patent B2

US 12,256,050 · App. 17/342,715 · Granted Mar 18, 2025

Inventor: Yoshinori Kimura (Tochigi, JP)

Assignee: CANON KABUSHIKI KAISHA

H04N13/111G06N3/045G06N3/08G06T7/97H04N13/282G06T2207/20081G06T2207/20084

Abstract

Claims (39)

1. An image processing apparatus comprising:

an image sensor including a plurality of pixels, with two light-receiving units only in each of the plurality of pixels, and an optical system that forms images of different viewpoints on the only two light-receiving units in each of the plurality of pixels; and

at least one processor or circuit configured to execute a plurality of tasks including:

an acquisition task configured to acquire (i) two first images made by capturing the same object at two viewpoints separated from each other in a first direction using the image sensor including the only two light-receiving units in each of the plurality of pixels and the optical system that forms the images of the different viewpoints on the only two light-receiving units in each of the plurality of pixels and (ii) an imaging condition associated with the two first images and including both of a focal length of the optical system and an ISO speed of the image sensor;

a feature amount generating task configured to generate two first feature amounts by inputting the two first images, the focal length of the optical system, and the ISO speed of the image sensor into a first neural network in a machine learning model;

a comparing task configured to generate a second feature amount by calculating corresponding points of the two first feature amounts through processing based on a matrix product of the two first feature amounts; and

an image processing task configured to estimate a second image at a viewpoint different from the two viewpoints by inputting the second feature amount into a second neural network in the machine learning model,

wherein the viewpoint different from the two viewpoints is shifted in a second direction orthogonal to the first direction, and

wherein, in the image processing task, an image at a viewpoint other than the two viewpoints is not input into the machine learning model in estimating the second image.

2. The image processing apparatus according to claim 1 , wherein the two first images are respectively input to two first neural networks having the same network architecture and weight as each other.

3. The image processing apparatus according to claim 1 , wherein the image processing task generates a third image that has an in-focus position changed from that of the second image.

4. An image processing method comprising:

an acquisition step of acquiring (i) two first images made by capturing the same object at two viewpoints separated from each other in a first direction using an image sensor including a plurality of pixels, with two light-receiving units only in each of the plurality of pixels, and an optical system that forms images of different viewpoints on the only two light-receiving units in each of the plurality of pixels and (ii) an imaging condition associated with the two first images and including both of a focal length of the optical system and an ISO speed of the image sensor;

a feature amount generating step of generating two first feature amounts by inputting the two first images, the focal length of the optical system, and the ISO speed of the image sensor into a first neural network in a machine learning model;

a comparing step of generating a second feature amount by calculating corresponding points of the two first feature amounts through processing based on a matrix product of the two first feature amounts; and

an image processing step of estimating a second image at a viewpoint different from the two viewpoints by inputting the second feature amount into a second neural network in the machine learning model,

wherein the viewpoint different from the two viewpoints is shifted in a second direction orthogonal to the first direction, and

wherein, in the image processing task, an image at a viewpoint other than the two viewpoints is not input into the machine learning model in estimating the second image.

5. A non-transitory computer-readable storage medium storing a computer program that causes a computer to execute the image processing method comprising:

an acquisition step of acquiring (i) two first images made by capturing the same object at two viewpoints separated from each other in a first direction using an image sensor including a plurality of pixels, with two light-receiving units only in each of the plurality of pixels, and an optical system that forms images of different viewpoints on the only two light-receiving units in each of the plurality of pixels and (ii) an imaging condition associated with the two first images and including both of a focal length of the optical system and an ISO speed of the image sensor;

a comparing step of generating a second feature amount by calculating corresponding points of the two first feature amounts through processing based on a matrix product of the two first feature amounts; and

an image processing step of estimating a second image at a viewpoint different from the two viewpoints by inputting the second feature amount into a second neural network in the machine learning model,

wherein the viewpoint different from the two viewpoints is shifted in a second direction orthogonal to the first direction, and

wherein, in the image processing task, an image at a viewpoint other than the two viewpoints is not input into the machine learning model in estimating the second image.

6. An image processing system including an image processing apparatus and a control apparatus that is capable of communicating with the image processing apparatus,

the image processing apparatus comprising:

an image sensor including a plurality of pixels, with two light-receiving units only in each of the plurality of pixels, and an optical system that forms images of different viewpoints on the only two light-receiving units in each of the plurality of pixels; and

at least one processor or circuit configured to execute a plurality of tasks including:

an acquisition task configured to acquire (i) two first images made by capturing the same object at two viewpoints separated from each other in a first direction using the image sensor including the only two light-receiving units in each of the plurality of pixels and the optical system that forms the images of the different viewpoints on the only two light-receiving units in each of the plurality of pixels and (ii) an imaging condition associated with the two first images and including both of a focal length of the optical system and an ISO speed of the image sensor;

a feature amount generating task configured to generate two first feature amounts by inputting the two first images, the focal length of the optical system, and the ISO speed of the image sensor into a first neural network in a machine learning model;

a comparing task configured to generate a second feature amount by calculating corresponding points of the two first feature amounts through processing based on a matrix product of the two first feature amounts; and

an image processing task configured to estimate a second image at a viewpoint different from the two viewpoints by inputting the second feature amount into a second neural network in the machine learning model,

wherein the viewpoint different from the two viewpoints is shifted in a second direction orthogonal to the first direction, and

wherein, in the image processing task, an image at a viewpoint other than the two viewpoints is not input into the machine learning model in estimating the second image;

wherein the control apparatus transmits a request for executing a process on the two first images, and

wherein the image processing apparatus executes the process on the two first images based on the request.

7. The image processing apparatus according to claim 1 ,

wherein the acquired imaging condition associated with the two first images further includes an aperture of the optical system and a pixel pitch of the image sensor, and

wherein the aperture of the optical system and the pixel pitch of the image sensor are input together with the two first images, the focal length of the optical system, and the ISO speed of the image sensor to the first neural network to generate the two first feature amounts.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (19)