Tehnički vjesnik, Vol. 30 No. 6, 2023.
Izvorni znanstveni članak
https://doi.org/10.17559/TV-20230701000778
Clustering Optimized Portrait Matting Algorithm Based on Improved Sparrow Algorithm
Xiang Wu
orcid.org/0000-0003-4280-5280
; College of Electrical Engineering, Henan University of Technology, Zhengzhou, China, 450000; Interdisciplinary Creative Application Research, GDPOLAR Co-work Space, Zhengzhou, China, 450000
*
Yuanhao Ma
orcid.org/0009-0007-3432-1047
; College of Electrical Engineering, Henan University of Technology, Zhengzhou, China, 450000
Hao Lian
orcid.org/0000-0003-2105-4983
; College of Electrical Engineering, Henan University of Technology, Zhengzhou, China, 450000
Xiang Fang
; College of Electrical Engineering, Henan University of Technology, Zhengzhou, China, 450000
Tianfei Chen
; Key Laboratory of Grain Information Processing and Control, Ministry of Education, Henan University of Technology, Zhengzhou, China, 450000
* Dopisni autor.
Sažetak
As a result of the influence of individual appearance and lighting conditions, aberrant noise spots cause significant mis-segmentation for frontal portraits. This paper presents an accurate portrait segmentation approach based on a combination of wavelet proportional shrinkage and an upgraded sparrow search (SSA) clustering algorithm to solve the accuracy challenge of segmentation for frontal portraits. The brightness component of the human portrait in HSV space is first subjected to wavelet scaling denoising. The elite inverse learning approach and adaptive weighting factor are then implemented to optimize the initial center location of the K-Means algorithm to improve the initial distribution and accelerate the convergence speed of SSA population members. The pixel segmentation accuracy of the proposed method is approximately 70% and 15% higher than two comparable traditional methods, while the similarity of color image features is approximately 10% higher. Experiments show that the proposed method has achieved a high level of accuracy in capricious lighting conditions.
Ključne riječi
elite backward learning strategy; K-means; portrait segmentation; sparrow search algorithm; wavelet proportional shrinkage
Hrčak ID:
309240
URI
Datum izdavanja:
25.10.2023.
Posjeta: 806 *