Universal approximated real-valued fast fourier transform for image blur detection

Xinghao Wang; Lu Bai; Yun Wu; Chongyan Gu

doi:10.1109/ISSC67739.2025.11291523

Universal approximated real-valued fast fourier transform for image blur detection

Xinghao Wang
, Lu Bai
, Yun Wu
, Chongyan Gu

School of Electronics, Electrical Engineering and Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

2 Downloads (Pure)

Abstract

This paper presents a primary study of universal approximation approach for the real-valued Fast Fourier Transform (FFT) with over-trimmed artificial neural network (ANN) for blur detection. By reducing the number of neurons in the hidden layer of ANN, we trade off the precision of the approximate real-valued FFT for computational complexity reduction. The fully connected ANN has only one hidden layer and no activate functions and bias, which simplifies the overall processing complexity compared to the normal ANN based FFT. Additional complexity reduction is achieved by dividing the image into sub-blocks. The performance of blur detection is verified, demonstrating the margin of approximate real-valued FFT with reduced computational complexity of up to ∼45% compared to untrimmed ANN, which is promising for deployment in resource-constrained edge devices.

Original language	English
Title of host publication	2025 36th Irish Signals and Systems Conference (ISSC): Proceedings
Publisher	IEEE
Number of pages	5
ISBN (Electronic)	9798331575939
ISBN (Print)	9798331575946
DOIs	https://doi.org/10.1109/ISSC67739.2025.11291523
Publication status	Published - 22 Dec 2025
Event	36th Irish Signals and Systems Conference - Letterkenny, Ireland Duration: 09 Jun 2025 → 10 Jun 2025

Publication series

Name	Proceedings of the Irish Systems and Signals Conference, ISSC
ISSN (Print)	2688-1446
ISSN (Electronic)	2688-1454

Conference

Conference	36th Irish Signals and Systems Conference
Country/Territory	Ireland
City	Letterkenny
Period	09/06/2025 → 10/06/2025

Publications and Copyright Policy

This work is licensed under Queen’s Research Publications and Copyright Policy.

Keywords

real-valued
image
blur detection

Access to Document

10.1109/ISSC67739.2025.11291523Licence: Unspecified

Universal approximated real-valued fast fourier transform for image blur detection
Copyright 2025 the authors. This is an accepted manuscript distributed under a Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium, provided the author and source are cited.
Accepted author manuscript, 9.02 MBLicence: CC BY

Cite this

@inproceedings{cad197a80ebf4a9781f291e21d01ee87,

title = "Universal approximated real-valued fast fourier transform for image blur detection",

abstract = "This paper presents a primary study of universal approximation approach for the real-valued Fast Fourier Transform (FFT) with over-trimmed artificial neural network (ANN) for blur detection. By reducing the number of neurons in the hidden layer of ANN, we trade off the precision of the approximate real-valued FFT for computational complexity reduction. The fully connected ANN has only one hidden layer and no activate functions and bias, which simplifies the overall processing complexity compared to the normal ANN based FFT. Additional complexity reduction is achieved by dividing the image into sub-blocks. The performance of blur detection is verified, demonstrating the margin of approximate real-valued FFT with reduced computational complexity of up to ∼45\% compared to untrimmed ANN, which is promising for deployment in resource-constrained edge devices. ",

keywords = "real-valued, image, blur detection",

author = "Xinghao Wang and Lu Bai and Yun Wu and Chongyan Gu",

year = "2025",

month = dec,

day = "22",

doi = "10.1109/ISSC67739.2025.11291523",

language = "English",

isbn = "9798331575946",

series = "Proceedings of the Irish Systems and Signals Conference, ISSC ",

publisher = "IEEE",

booktitle = "2025 36th Irish Signals and Systems Conference (ISSC): Proceedings",

note = "36th Irish Signals and Systems Conference ; Conference date: 09-06-2025 Through 10-06-2025",

}

Wang, X, Bai, L , Wu, Y & Gu, C 2025, Universal approximated real-valued fast fourier transform for image blur detection. in 2025 36th Irish Signals and Systems Conference (ISSC): Proceedings. Proceedings of the Irish Systems and Signals Conference, ISSC , IEEE, 36th Irish Signals and Systems Conference, Letterkenny, Ireland, 09/06/2025. https://doi.org/10.1109/ISSC67739.2025.11291523

TY - GEN

T1 - Universal approximated real-valued fast fourier transform for image blur detection

AU - Wang, Xinghao

AU - Bai, Lu

AU - Wu, Yun

AU - Gu, Chongyan

PY - 2025/12/22

Y1 - 2025/12/22

N2 - This paper presents a primary study of universal approximation approach for the real-valued Fast Fourier Transform (FFT) with over-trimmed artificial neural network (ANN) for blur detection. By reducing the number of neurons in the hidden layer of ANN, we trade off the precision of the approximate real-valued FFT for computational complexity reduction. The fully connected ANN has only one hidden layer and no activate functions and bias, which simplifies the overall processing complexity compared to the normal ANN based FFT. Additional complexity reduction is achieved by dividing the image into sub-blocks. The performance of blur detection is verified, demonstrating the margin of approximate real-valued FFT with reduced computational complexity of up to ∼45% compared to untrimmed ANN, which is promising for deployment in resource-constrained edge devices.

AB - This paper presents a primary study of universal approximation approach for the real-valued Fast Fourier Transform (FFT) with over-trimmed artificial neural network (ANN) for blur detection. By reducing the number of neurons in the hidden layer of ANN, we trade off the precision of the approximate real-valued FFT for computational complexity reduction. The fully connected ANN has only one hidden layer and no activate functions and bias, which simplifies the overall processing complexity compared to the normal ANN based FFT. Additional complexity reduction is achieved by dividing the image into sub-blocks. The performance of blur detection is verified, demonstrating the margin of approximate real-valued FFT with reduced computational complexity of up to ∼45% compared to untrimmed ANN, which is promising for deployment in resource-constrained edge devices.

KW - real-valued

KW - image

KW - blur detection

U2 - 10.1109/ISSC67739.2025.11291523

DO - 10.1109/ISSC67739.2025.11291523

M3 - Conference contribution

SN - 9798331575946

T3 - Proceedings of the Irish Systems and Signals Conference, ISSC

BT - 2025 36th Irish Signals and Systems Conference (ISSC): Proceedings

PB - IEEE

T2 - 36th Irish Signals and Systems Conference

Y2 - 9 June 2025 through 10 June 2025

ER -

Universal approximated real-valued fast fourier transform for image blur detection

Abstract

Publication series

Conference

Publications and Copyright Policy

Keywords

Access to Document

Fingerprint

Cite this