Granted Patent B2

US 12,608,880 · App. 18/161,780 · Granted Apr 21, 2026

Inventors: Hae-Gon Jeon (Gwangju, KR); Hyunjun Jung (Gwangju, KR)

Assignee: GIST (Gwangju Institute of Science and Technology)

G06T15/205G06N3/0475G06T3/18G06T3/4046G06T3/4053G06T5/77G06T7/557G06T2207/10052G06T2207/20084G06T2207/20228

Abstract

Claims (27)

1 . A view extrapolation method using an epipolar plane image, the view extrapolation method comprising:

creating an Epipolar Plane Image (EPI) using a target image and a depth map corresponding to the target image;

creating a super-resolution EPI and a disparity of the EPI on the basis of the EPI;

creating an EPI mask by warping the super-resolution EPI in accordance with the disparity; and

creating a restored EPI by applying the EPI mask to the warped super-resolution EPI, and creating an extrapolated image for the target image using the restored EPI,

wherein the creating of an EPI mask includes creating an EPI mask of which a value is determined by a coordinate of a pixel that is propagated in accordance with the disparity when the super-resolution EPI is warped,

wherein the creating of a restored EPI includes:

creating a masked EPI by performing element-wise multiplication on the EPI mask and the warped super-resolution EPI; and

creating the restored EPI by restoring the masked EPI, and

wherein the creating of a restored EPI includes creating the restored EPI by inputting the masked EPI to a Generative Adversarial Network (GAN).

2 . The view extrapolation method of claim 1 ,

wherein the creating of an EPI comprises:

creating a plurality of stacked sub-aperture images through warping according to the depth map; and

creating the EPI from the plurality of stacked sub-aperture images by axially cutting the plurality of stacked sub-aperture images.

3 . The view extrapolation method of claim 1 , wherein the creating of an EPI includes:

receiving the target image taken at a first view and a reference image taken at a second view; and

creating the depth map corresponding to the target image on the basis of the target image and the reference image.

4 . The view extrapolation method of claim 1 , wherein the creating of an EPI includes creating the EPI by applying Gaussian blur to the depth map.

5 . The view extrapolation method of claim 1 , wherein the creating of a disparity includes creating the disparity on the basis of an inclination of the EPI.

6 . The view extrapolation method of claim 1 , wherein the warping of the super-resolution EPI includes warping pixels of each of lines constituting the super-resolution EPI in accordance with the disparity of each of the lines.

7 . The view extrapolation method of claim 6 , wherein the warping of the super-resolution EPI includes:

determining a pixel value of a first line of the super-resolution EPI as a pixel value of a first line of the warped super-resolution EPI; and

determining the warped super-resolution EPI by propagating pixel values of each of lines of the warped super-resolution EPI to next lines in accordance with the disparity of each of the lines.

8 . The view extrapolation method of claim 1 , wherein the creating of an EPI mask includes creating a binary EPI mask in which the value of a coordinate of a pixel that is propagated in accordance with the disparity is 1 and values of coordinates of the other pixels are 0.

9 . The view extrapolation method of claim 1 , wherein the creating of an extrapolated image includes:

creating a plurality of sub-aperture images using the restored EPI; and

determining any one of the plurality of sub-aperture images as the extrapolated image.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (14)