Deep CNN Based Interpolation Filter for High Efficiency Video Coding

Title

Deep CNN Based Interpolation Filter for High Efficiency Video Coding

Creator

Joy H.K.; Kounte M.R.

Description

Video coding is a current focus in research area as the world focus more on multimedia transfer. High Efficiency Video Coding (HECV) is prominent among existing one. The interpolation in HEVC with fixed half-pel interpolation filter uses fixed interpolation filter derived from traditional signal processing methods. Some research came up with CNN based interpolation filter too, here we are proposing a deep learning-based interpolation filter to perform interpolation in inter prediction in HEVC. The network extracts the low-resolution image and extract the patch and feature in that to predict a high-resolution image. The network is trained to predict the HR image for the given patch, it can be repeated to generate the full frame in the HEVC. The system uses cleave approach to reduce the computational complexity. The trained network is validated and tested for different inputs. The results show an improvement of 2.38% in BD-bitrate saving for low delay configuration. 2024 IEEE.

Source

2nd International Conference on Intelligent Data Communication Technologies and Internet of Things, IDCIoT 2024, pp. 519-524.

Date

2024-01-01

Publisher

Institute of Electrical and Electronics Engineers Inc.

Subject

Deep CNN; HEVC; MV; SRCNN

Coverage

Joy H.K., CHRIST University, Department of Computer Science, Bangaluru, India; Kounte M.R., School of ECE, REVA University, Bangaluru, India

Rights

Restricted Access

Relation

ISBN: 979-835032753-3

Format

Online

Language

English

Type

Conference paper

Identifier

https://doi.org/10.1109/IDCIoT59759.2024.10467280

https://www.scopus.com/inward/record.uri?eid=2-s2.0-85190119176&doi=10.1109%2fIDCIoT59759.2024.10467280&partnerID=40&md5=9be0cee7f27834e4cadffae41d5f503f