Publications & Patents

Academic Publications

Enhancing HEVC Compressed Videos with a Partition-Masked Convolutional Neural Network

He, Xiaoyi, et al.

International Conference on Image Processing (ICIP)2018(Oral presentation)
98 citations

Abstract

We propose a novel partition-masked convolutional neural network (CNN) that utilizes partition information in video encoder to enhance compressed videos. Achieved approximately 10% bitrate savings on benchmark sequences.

Partition-aware Adaptive Switching Networks for Post-processing in HEVC

Lin, Weiyao, He, Xiaoyi, et al.

IEEE Transactions on Multimedia2020
61 citations

Abstract

We introduce an adaptive-switching neural network (ASN) that consists of multiple independent CNNs to adaptively handle the variations in content and distortion within compressed video frames, further reducing visual artifacts.

Key-point Sequence Lossless Compression for Intelligent Video Analysis

Lin, Weiyao, He, Xiaoyi, et al.

IEEE MultiMedia2020
23 citations

Abstract

This paper presents a lossless compression method for key-point sequences in videos to enable efficient intelligent video analysis.

A Multimodal Lossless Coding Method for Skeletons in Videos

He, Xiaoyi, et al.

IEEE International Conference on Multimedia & Expo Workshop (ICME)2019
2 citations

Abstract

This paper presents a lossless compression method for skeleton data in videos based on spatial and temporal correlation, achieving about 84% compression ratio on test surveillance sequences.

Adaptive Lossless Compression of Skeleton Sequences

Lin, Weiyao, Shinde, TS, Dai, Wenrui, Liu, Ming, He, Xiaoyi, et al.

Signal Processing: Image Communication2020
2 citations

Abstract

An adaptive compression framework for skeleton data that dynamically selects optimal coding methods based on data characteristics.

Patents

Image Illumination Compensation Method and System Based on Tone Shift Estimation and Point-by-Point Tone Mapping

He, Xiaoyi; et al.

Patent CN106504200BGranted2022

Abstract

An image illumination compensation method based on tone shift estimation and point-by-point tone mapping. The system extracts tone shift and mapping functions from calibration charts to perform illumination compensation on images, providing superior results compared to traditional gamma correction methods.

Convolutional Neural Network-based In-Loop Filtering Method and System for Video Coding and Decoding

Lin, Weiyao; He, Xiaoyi

Patent CN108134932BGranted2021

Abstract

A method and system for in-loop filtering in video coding and decoding based on convolutional neural networks. The method uses CNN models to enhance the quality of compressed video frames within the coding loop, reducing compression artifacts while maintaining coding efficiency.

Generative Adversarial Network-based In-Loop Filtering Method and System for Video Coding and Decoding

Lin, Weiyao; He, Xiaoyi

Patent CN108174225BGranted2021

Abstract

A method and system for in-loop filtering in video coding and decoding based on generative adversarial networks (GAN). The approach uses a GAN architecture with generator and discriminator models to improve the quality of reconstructed video frames within the coding loop, providing better visual quality than traditional CNN methods.

In-Loop Filtering Implementation Method Based on Joint Construction and Adaptive Selection of Multiple Networks

Lin, Weiyao; He, Xiaoyi; et al.

Patent CN108520505BGranted2021

Abstract

A method for implementing in-loop filtering in video coding based on joint construction and adaptive selection of multiple neural networks. The system dynamically selects the optimal filtering network based on content characteristics, offering improved video quality with adaptive processing.