公开(公告)号：US20230196520A1

公开(公告)日：2023-06-22

申请号：US17974383

申请日：2022-10-26

Applicant: POSTECH Research and Business Development Foundation

Inventor： Seung Yong LEE ,  Hyeong Seok SON ,  Sung Hyun CHO ,  Jun Yong LEE

IPC: G06T5/00 ,  G06N3/02

CPC classification number: G06T5/003 ,  G06N3/02

Abstract: The present disclosure provides a method of effectively deblurring a defocus blur in an input image based on an inverse kernel. The defocus deblurring method includes: generating, by an encoder network, an input feature map by encoding the input image; filtering, by an atrous convolution network including a plurality of atrous convolutional layers arranged in parallel, the input feature map to generate an output feature map having reduced blur component; and generating, by a decoder network, an output image having reduced blur from the output feature map with the reduced blur component and the input image.

公开(公告)号：US20220366539A1

公开(公告)日：2022-11-17

申请号：US17770993

申请日：2020-11-11

Applicant: POSTECH RESEARCH AND BUSINESS DEVELOPMENT FOUNDATION

Inventor： Seung Yong LEE ,  Sung Hyun CHO ,  Hyeong Seok SON

IPC: G06T5/00 ,  G06T5/50 ,  G06N3/04

Abstract: An image processing method and apparatus based on machine learning are disclosed. The image processing method based on machine learning, according to the present invention, may comprise the steps of: generating a first corrected image by inputting an input image to a first convolution neural network; generating an intermediate image on the basis of the input image; performing machine learning on a first loss function of the first convolution neural network on the basis of the first corrected image and the intermediate image; and performing machine learning on a second loss function of the first convolution neural network on the basis of the first corrected image and a natural image.

公开(公告)号：US20220198616A1

公开(公告)日：2022-06-23

申请号：US17497824

申请日：2021-10-08

Applicant: POSTECH Research and Business Development Foundation

Inventor： Seung Yong LEE ,  Jun Yong LEE ,  Hyeong Seok SON ,  Sung Hyun CHO

IPC: G06T5/00 ,  G06T5/50 ,  G06T7/246 ,  G06T7/33 ,  G06N3/04

Abstract: A video quality improvement method may comprise: inputting a structure feature map converted from current target frame by first convolution layer to first multi-task unit and second multi-task unit, which is connected to an output side of first multi-task unit, among the plurality of multi-task units; inputting a main input obtained by adding the structure feature map to a feature space, which is converted by second convolution layer from those obtained by concatenating, in channel dimension, a previous target frame and a correction frame of the previous frame to first multi-task unit; and inputting current target frame to Nth multi-task unit connected to an end of output side of second multi-task unit, wherein Nth multi-task unit outputs a correction frame of current target frame, and machine learning of the video quality improvement model is performed using an objective function calculated through the correction frame of current target frame.