Granted Patent B2

US 12,563,311 · App. 18/302,537 · Granted Feb 24, 2026

Inventor: Dong Ik Kim (Icheon-si, KR)

Assignee: SK hynic Inc.

H04N25/41H04N25/46H04N25/704

Abstract

Claims (50)

1 . An image processing device comprising:

a memory configured to store a super resolution network trained to output a corrected image based on an input image; and

a processor configured to output a super resolution image based on a first image acquired by a first image sensor and a corrected image output from the super resolution network,

wherein the super resolution network is a model trained by using second images acquired through a second image sensor,

wherein a type of the second image sensor is different from a type of the first image sensor,

wherein the first image sensor includes only sets of phase detection pixels, each of the set includes a first microlens and photodiodes disposed under the first microlens,

wherein the first microlens is shared with phase detection pixels included in one of the sets, and

wherein the second image sensor includes a plurality of normal pixels, each of the plurality of normal pixels includes a second micro lens and a photodiode.

2 . The image processing device of claim 1 , wherein the first microlens refracts light and the photodiodes sense intensities of light refracted through the first microlens, and

wherein the first microlens has a size for covering the photodiodes which include a photodiode and at least one photodiode adjacent to the photodiode.

3 . The image processing device of claim 1 , wherein the second microlens refracts light and the photodiode senses an intensity of light refracted through the second microlens, and

wherein the second microlens has a size for covering the photodiode.

4 . The image processing device of claim 1 , wherein the processor acquires a binning image obtained by performing binning processing on pixel values included in the first image, acquires an upscale image for the binning image, and acquires the super resolution image by using the super resolution network.

5 . The image processing device of claim 4 , wherein the first image includes a plurality of unit regions, and each of the plurality of unit regions includes pixel values of the same color, and

wherein the processor includes a binning processor configured to acquire the binning image including a plurality of average pixel values corresponding to the plurality of unit regions by performing an average operation on the pixel values included in each of the plurality of unit regions.

6 . The image processing device of claim 4 , wherein the processor includes a scaler configured to acquire the upscale image by performing interpolation processing on the binning image.

7 . The image processing device of claim 4 , wherein the super resolution network includes a plurality of filters, and

wherein the processor includes a super resolution image acquirer configured to acquire the corrected image by performing a convolution operation, using the upscale image and each of the plurality of filters, and acquire the super resolution image by summing the upscale image and the corrected image.

8 . The image processing device of claim 7 , wherein a plurality of filters of the super resolution network are updated based on a loss value between a binning image corresponding to a selected second image among the second images, and an output image obtained by summing a preprocessing image and a residual image corresponding to the selected second image,

wherein the preprocessing image is an image obtained by performing downscaling processing and upscaling processing, on the binning image, and

wherein the residual image is an image output from the super resolution network in which the preprocessing image is input.

9 . The image processing device of claim 1 , wherein the first image and the second image include pixel values of a red color, pixel values of a green color, and pixel values of a blue color, which are arranged according to a quad Bayer pattern, and

wherein the super resolution image includes pixel values of a red color, pixel values of a green color, and pixel values of a blue color, which are arranged according to a quad Bayer pattern.

10 . An image processing method comprising:

acquiring a first image through a first image sensor; and

acquiring a super resolution image based on the first image through a super resolution network trained using second images acquired by a second image sensor,

wherein a type of the second image sensor is different from a type of the first image sensor,

wherein the first image sensor includes only sets of phase detection pixels, each of the set includes a first microlens and photodiodes disposed under the first microlens,

wherein the first microlens is shared with phase detection pixels included in one of the sets, and

wherein the second image sensor includes a plurality of normal pixels, each of the plurality of normal pixels includes a second micro lens and a photodiode.

11 . The image processing method of claim 10 , further comprising acquiring a binning image including a plurality of average pixel values corresponding to a plurality of unit regions included in the first image by performing an average operation on pixel values included in each of the plurality of unit regions.

12 . The image processing method of claim 11 , further comprising acquiring an upscale image by performing interpolation processing on the binning image.

13 . The image processing method of claim 12 , wherein the acquiring of the super resolution image includes:

acquiring a corrected image by performing a convolution operation, using the upscale image and the super resolution network; and

acquiring the super resolution image by summing the upscale image and the corrected image.

14 . The image processing method of claim 10 , further comprising training a plurality of filters included in the super resolution network by using the second images acquired through the second image sensor including a plurality of microlenses each having a size for covering a plurality of photodiodes.

15 . The image processing method of claim 14 , wherein the training of the plurality of filters includes:

acquiring a preprocessing image for a selected second image, among the second images;

acquiring a residual image by performing a convolution operation based on the preprocessing image and each of the plurality of filters;

acquiring an output image by summing the preprocessing image and the residual image; and

updating the plurality of filters based on a loss value of the output image and a binning image for the selected second image.

16 . The image processing method of claim 15 , wherein the acquiring of the preprocessing image includes:

acquiring the binning image by performing binning processing on the selected second image;

acquiring a downscale image by performing downscaling processing on the binning image; and

acquiring, as the preprocessing image, an upscale image by performing upscaling processing on the downscale image.

17 . The image processing method of claim 10 , further comprising receiving and storing the super resolution network trained in an external device.

18 . The image processing method of claim 10 ,

wherein a size of the first microlens is greater than a size of the second microlenses microlens.

19 . The image processing method of claim 10 , wherein the first image and the second image include pixel values of a red color, pixel values of a green color, and pixel values of a blue color, which are arranged according to a quad Bayer pattern, and

wherein the super resolution image includes pixel values of a red color, pixel values of a green color, and pixel values of a blue color, which are arranged according to a quad Bayer pattern.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (13)