notesum.ai

Published at December 10

Motion Artifact Removal in Pixel-Frequency Domain via Alternate Masks and Diffusion Model

eess.IV

cs.CV

Released Date: December 10, 2024

Authors: Jiahua Xu¹, Dawei Zhou, Lei Hu, Jianfeng Guo, Feng Yang, Zaiyi Liu, Nannan Wang, Xinbo Gao

Aff.: ¹State Key Laboratory of Integrated Services Networks, Xidian University, Xi'an, China

Arxiv: http://arxiv.org/pdf/2412.07590v1

Dataset	Methods	PSNR $\uparrow$	SSIM $\uparrow$	Lpips $\downarrow$	FSIM $\uparrow$	VIF $\uparrow$	VSI $\uparrow$	GMSD $\downarrow$
Brain Random	Input	25.97^∗∗	0.685^∗∗	0.349^∗∗	0.812^∗∗	0.295^∗∗	0.945^∗∗	0.114^∗∗
	Pix2pix	26.76^∗∗	0.841	0.151^∗∗	0.882^∗∗	0.392^∗∗	0.966^∗∗	0.065^∗∗
	CycleGAN	25.94^∗∗	0.719^∗∗	0.277^∗∗	0.846^∗∗	0.338^∗∗	0.955^∗∗	0.091^∗∗
	UDDN	26.07^∗∗	0.823^∗∗	0.163^∗∗	0.879^∗∗	0.372^∗∗	0.965^∗∗	0.065^∗∗
	DR2	26.60^∗∗	0.780^∗∗	0.174^∗∗	0.859^∗∗	0.372^∗∗	0.960^∗∗	0.083^∗∗
	GDP	26.08^∗∗	0.739^∗∗	0.193^∗∗	0.847^∗∗	0.313^∗∗	0.957^∗∗	0.092^∗∗
	PFAD (ours)	27.60	0.829	0.147	0.896	0.426	0.970	0.061
Knee Random	Input	29.05^∗∗	0.660^∗∗	0.340^∗∗	0.846^∗∗	0.266^∗∗	0.965^∗∗	0.110^∗∗
	Pix2pix	29.37^∗∗	0.705^∗	0.279^∗∗	0.861^∗∗	0.296^∗∗	0.970^∗∗	0.085^∗∗
	CycleGAN	27.34^∗∗	0.663^∗∗	0.289^∗∗	0.849^∗∗	0.276^∗∗	0.967^∗∗	0.095^∗∗
	UDDN	28.50^∗∗	0.653^∗∗	0.270^∗∗	0.852^∗∗	0.274^∗∗	0.967^∗∗	0.089^∗∗
	DR2	29.23^∗∗	0.686^∗∗	0.333^∗∗	0.835^∗∗	0.300^∗∗	0.964^∗∗	0.101^∗∗
	GDP	28.08^∗∗	0.599^∗∗	0.285^∗∗	0.840^∗∗	0.213^∗∗	0.966^∗∗	0.104^∗∗
	PFAD (ours)	29.64	0.712	0.220	0.869	0.314	0.972	0.080
Abdominal Respiratory	Input	35.37^∗∗	0.816^∗∗	0.234^∗∗	0.944^∗∗	0.441^∗∗	0.986^∗∗	0.064^∗∗
	Pix2pix	36.14^∗∗	0.902^∗∗	0.147^∗∗	0.955^∗∗	0.481^∗∗	0.988^∗∗	0.058^∗∗
	CycleGAN	34.68^∗∗	0.902^∗∗	0.136^∗∗	0.949^∗∗	0.468^∗∗	0.986^∗∗	0.067^∗∗
	UDDN	35.14^∗∗	0.896^∗∗	0.117^∗∗	0.951^∗∗	0.473^∗∗	0.986^∗∗	0.066^∗∗
	DR2	34.37^∗∗	0.794^∗∗	0.168^∗∗	0.933^∗∗	0.389^∗∗	0.983^∗∗	0.076^∗∗
	GDP	33.41^∗∗	0.760^∗∗	0.220^∗∗	0.919^∗∗	0.320^∗∗	0.981^∗∗	0.083^∗∗
	PFAD (ours)	37.24	0.918	0.103	0.962	0.544	0.990	0.051