Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
METHOD FOR REMOVING COMPRESSION DISTORTION BY USING CNN
Document Type and Number:
WIPO Patent Application WO/2019/112084
Kind Code:
A1
Abstract:
Embodiments relate to a method for removing compression distortion from a compressed image, wherein images are classified into one or more categories according to the degree of compression distortion, and thus compression distortion of an input image is removed as a corresponding group according to the degree of distortion of the input image. When the images are classified into categories according to the degree of compression distortion, appropriate compression distortion for each category can be performed by using QP information indicating the degree of compression of an image, and configuring a network, which removes compression distortion for each group, and learning the same, thereby enabling conventional compression distortion performance to be improved. In addition, the distortion of all compressed images can be removed regardless of the degree of compression distortion through a determination network capable of identifying an image according to the compression distortion.

Inventors:
KIM MUN CHURL (KR)
JEON JIN (KR)
Application Number:
PCT/KR2017/014263
Publication Date:
June 13, 2019
Filing Date:
December 06, 2017
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
KOREA ADVANCED INST SCI & TECH (KR)
International Classes:
H04N19/85; G06N3/08; H04N19/124; H04N19/176
Foreign References:
KR20170003476A2017-01-09
KR20110049570A2011-05-12
KR20160141201A2016-12-08
KR20160131071A2016-11-15
Other References:
PARK, WOON SUNG ET AL.: "CNN (Convolutional Neural Network) Based In-loop Filter in HEVC", THE KOREAN INSTITUTE OF BROADCAST AND MEDIA ENGINEERS SUMMER CONFERENCE, 1 July 2016 (2016-07-01), pages 418 - 421
Attorney, Agent or Firm:
MUHANN PATENT & LAW FIRM (KR)
Download PDF: