IP Library › Granted Patent US 12,641,225
Granted Patent B2
US 12,641,225 · App. 18/579,854 · Granted May 26, 2026

Image encoding/decoding method using neural network-based in-loop filter, device, and recording medium storing bitstream

Inventors: Hyun Gyu Kim (Seoul, KR); Dae Yeon Kim (Seoul, KR); Yung-Lyul Lee (Seoul, KR); Yang Woo Kim (Seoul, KR); Myung Jun Kim (Seoul, KR); Nam Uk Kim (Seoul, KR)
Assignee: INDUSTRY ACADEMY COOPERATION FOUNDATION OF SEJONG UNIVERSITY
H04N19/117G06T5/20G06V10/44H04N19/124H04N19/186H04N19/82
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,641,225
App. No.
18/579,854
Granted
May 26, 2026
Kind
B2
Abstract

An image decoding method using a neural network-based in-loop filter may comprise obtaining a first image feature from an input image, obtaining a block information feature of the input image from block information of the input image, obtaining a second image feature by removing noise and distortion of the first image feature based on the block information feature, and reconstructing the input image based on the second image feature. The block information may comprise at least one of a block boundary map indicating a block partition structure of the input image or a block distribution map indicating encoding information of the input image.

Claims (43)

1 . An image decoding method using a neural network-based in-loop filter, comprising:

obtaining a first image feature from an input image;

obtaining a block information feature of the input image from block information of the input image;

obtaining a second image feature based on the first image feature and the block information feature; and

reconstructing the input image based on the second image feature,

wherein the block information comprises at least one of a block boundary map indicating a block partition structure of the input image or a block distribution map indicating encoding information of the input image,

wherein the second image feature is obtained by adding the block information feature to the first image feature,

wherein the block information feature is transformed into a two-dimensional map and then is added to the first image feature in channel units.

2 . The image decoding method of claim 1 ,

wherein the block boundary map and the block distribution map are grouped and reorganized into a block map group, and

wherein the block information feature is obtained from the block map group.

3 . The image decoding method of claim 1 ,

wherein the block boundary map and the block distribution map are merged and reorganized into a single block information map, and

wherein the block information feature is obtained from the block information map.

4 . The image decoding method of claim 3 , wherein the block information map is constructed by a weighted sum of sample values of corresponding pixels in the block boundary map and the block distribution map.

5 . The image decoding method of claim 1 , wherein the encoding information comprises at least one of prediction information or transform information of the input image.

6 . The image decoding method of claim 1 , wherein the first image feature is individually obtained with respect to each of a luma component and chroma component of the input image.

7 . The image decoding method of claim 1 , further comprising applying a channel attention based on a quantization parameter of the input image to the block information,

wherein the block information feature is obtained from the block information to which the channel attention is applied.

8 . An image encoding method using a neural network-based in-loop filter, comprising:

obtaining a first image feature from an input image;

obtaining a block information feature of the input image from block information of the input image;

obtaining a second image feature based on the first image feature and the block information feature; and

reconstructing the input image based on the second image feature,

wherein the block information comprises at least one of a block boundary map indicating a block partition structure of the input image or a block distribution map indicating encoding information of the input image,

wherein the second image feature is obtained by adding the block information feature to the first image feature,

wherein the block information feature is transformed into a two-dimensional map and then is added to the first image feature in channel units.

9 . The image encoding method of claim 8 , wherein the block boundary map and the block distribution map are grouped and reorganized into a block map group, and

wherein the block information feature is obtained from the block map group.

10 . The image encoding method of claim 8 ,

wherein the block boundary map and the block distribution map are merged and reorganized into a single block information map, and

wherein the block information feature is obtained from the block information map.

11 . The image encoding method of claim 10 , wherein the block information map is constructed by a weighted sum of sample values of corresponding pixels in the block boundary map and the block distribution map.

12 . The image encoding method of claim 8 , wherein the first image feature is individually obtained with respect to each of a luma component and chroma component of the input image.

13 . The image encoding method of claim 8 , further comprising applying a channel attention based on a quantization parameter of the input image to the block information, wherein the block information feature is obtained from the block information to which the channel attention is applied.

14 . A non-transitory computer-readable recording medium for storing a bitstream generated by an image encoding method using a neural network-based in-loop filter, comprising instructions stored thereon, that when executed on a processor, performs the image encoding method comprising:

obtaining a first image feature from an input image;

obtaining a block information feature of the input image from block information of the input image;

obtaining a second image feature based on the first image feature and the block information feature; and

reconstructing the input image based on the second image feature,

wherein the block information comprises at least one of a block boundary map indicating a block partition structure of the input image or a block distribution map indicating encoding information of the input image,

wherein the second image feature is obtained by adding the block information feature to the first image feature,

wherein the block information feature is transformed into a two-dimensional map and then is added to the first image feature in channel units.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jan 17, 2024
From: KIM, HYUN GYU; KIM, DAE YEON; LEE, YUNG-LYUL; KIM, YANG WOO; KIM, MYUNG JUN; KIM, NAM UK
To: INDUSTRY ACADEMY COOPERATION FOUNDATION OF SEJONG UNIVERSITY
Reel/Frame 066146/0154 →
Priority Claims (1)
KR 10-2021-0092720 · Jul 15, 2021 · national
Continuity (1)
Related Publication 20240333923A1 · Oct 3, 2024
References Cited (12)
US 11677948B2 · Besenbruch · 2023 [cited by examiner]
US 20200154145A1 · Du · 2020 [cited by examiner]
US 20210021823A1 · Na · 2021 [cited by examiner]
US 20210160522A1 · Lee · 2021 [cited by examiner]
US 20220375602A1 · Jaber · 2022 [cited by examiner]
JP 2020185390A · 2020 [cited by applicant]
KR 1020200000548A · 2020 [cited by applicant]
KR 1020200040773A · 2020 [cited by applicant]
WO 2019009452A1 · 2019 [cited by applicant]
Wang. “Attention-Based Dual-Scale CNN In-Loop Filter for Versatile Video Coding”. Sep.-Oct. 2019. (Year: 2019). [cited by examiner]
International Search Report for PCT/KR2022/009818 by Korean Intellectual Property Office dated Oct. 7, 2022. [cited by applicant]
Kim, Yangwoo et al. “CNN-based In-loop Filtering Using Block Information”, 2019 Korean Society of Broadcasting and Media Engineering Fall Conference, pp. 27-29. [cited by applicant]