US 11,985,358 B2
Artifact removal method and apparatus based on machine learning, and method and apparatus for training artifact removal model based on machine learning
Yan Shang, Shenzhen (CN)
Assigned to TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED, Shenzhen (CN)
Filed by TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED, Guangdong (CN)
Filed on Oct. 14, 2021, as Appl. No. 17/501,217.
Application 17/501,217 is a continuation of application No. PCT/CN2020/120006, filed on Oct. 9, 2020.
Claims priority of application No. 201910984591.3 (CN), filed on Oct. 16, 2019.
Prior Publication US 2022/0038749 A1, Feb. 3, 2022
Int. Cl. H04N 19/86 (2014.01); G06F 18/214 (2023.01); G06F 18/25 (2023.01); G06N 20/00 (2019.01); H04N 19/117 (2014.01)
CPC H04N 19/86 (2014.11) [G06F 18/214 (2023.01); G06F 18/253 (2023.01); G06N 20/00 (2019.01); H04N 19/117 (2014.11)] 16 Claims
OG exemplary drawing
 
1. An artifact removal method based on machine learning, performed by an electronic device, the method comprising:
invoking an artifact removal model to perform feature extraction on an ith original image frame of a video, to obtain a feature vector of the ith original image frame, the video comprising at least two original image frames and i being a positive integer;
invoking the artifact removal model to perform dimension reduction on the feature vector, to obtain a feature vector after the dimension reduction;
invoking the artifact removal model to perform feature reconstruction on the feature vector after the dimension reduction, to predict a residual between the ith original image frame and an image frame obtained after the ith original image frame is compressed;
invoking the artifact removal model to add the predicted residual and the ith original image frame, to obtain a target image frame after artifact removal processing; and
sequentially performing encoding and compression on at least two target image frames corresponding to the at least two original image frames, to obtain a video frame sequence after artifact removal,
wherein the artifact removal model comprises a first 1×1 convolutional layer, a second 1×1 convolutional layer, a first feature extraction layer, and a feature fusion layer, and
wherein the invoking the artifact removal model to perform the dimension reduction on the feature vector comprises:
invoking the first 1×1 convolutional layer to perform the dimension reduction on the feature vector, to obtain a first feature vector after the dimension reduction;
invoking the second 1×1 convolutional layer to perform the dimension reduction on the feature vector, to obtain a second feature vector after the dimension reduction;
invoking the first feature extraction layer to perform feature extraction on the second feature vector, to obtain a second feature vector after the feature extraction; and
invoking the feature fusion layer to fuse the first feature vector with the second feature vector after the feature extraction, to obtain the feature vector after the dimension reduction.