A small-object segmentation algorithm for intercluster imbalance based on histogram- and stickiness-aware boosting

Traditional fuzzy C-means (FCM) clustering and its variants, as important unsupervised image segmentation methods, have average performance, usually perform poorly in the face of unbalanced datasets, and are sensitive to the initial position. Therefore, in this paper, we propose a small-object segme...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Engineering applications of artificial intelligence Ročník 163; s. 113045
Hlavní autori: Chong, Qianpeng, Wen, Jiakun, Wan, Qianhui, Zeng, Wenyi, Yin, Qian, Cheng, Dong
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Elsevier Ltd 01.01.2026
Predmet:
Fuzzy C-means algorithm
Image segmentation
Small-object detection
Unequal clusters
Image segmentation
Unequal clusters
Fuzzy C-means algorithm
Small-object detection
ISSN:0952-1976
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Traditional fuzzy C-means (FCM) clustering and its variants, as important unsupervised image segmentation methods, have average performance, usually perform poorly in the face of unbalanced datasets, and are sensitive to the initial position. Therefore, in this paper, we propose a small-object segmentation FCM algorithm for intercluster imbalance based on histograms and stickiness-aware boosting. This algorithm has three main parts: (1) histogram boosting of images, achieved by introducing histogram boosting factors to balance the contributions of different grayscale samples; (2) boosting factor selection, guided by connectivity region information and separation distances; and (3) FCM clustering and image segmentation based on the selected boosting factor. The separation distances are innovatively combined with absolute distances, relative distances, and stickiness of pixels to clusters. This approach allows the state where the sum of the separation distances is minimized to effectively represent the exact state of the small-object segmentation. The experimental results show that the proposed algorithm has the characteristics of high accuracy, fast speed, and good stability in small-object detection. In some challenging scenarios and when the target categories are unbalanced, the segmentation accuracy of the proposed algorithm reaches 99.07%, whereas the normalized mutual information, F1 score, and mean intersection over union reach 97.65%, 95.47%, and 90.31%, respectively. Our resource code can be accessed at https://github.com/wenxiaomo/HBFCM.
ISSN:0952-1976
DOI:10.1016/j.engappai.2025.113045