Image Segmentation Algorithm Based on Anisotropic Fuzzy Local Convolution
Year: 2026
Type: conference-paper
Abstract: - segmentation of images with thin elongated structures remains difficult when the target object has a branching shape, small thickness, weak contrast, and blurred boundaries. In such conditions, conventional local processing may lose faint fragments, interrupt structural continuity, and distort the extracted region. This paper presents an image segmentation algorithm based on anisotropic fuzzy local convolution. The method forms a local aggregation layer where the contribution of neighboring points is controlled by the directional geometry of the neighborhood and by fuzzy agreement with the structure being extracted. This design keeps the model sensitive to elongated fragments and makes its response more stable in uncertain image regions. The paper describes the mathematical model of the local layer, its use in a multilayer segmentation scheme, the construction of the output probability map, and the training procedure based on a combined loss function. The experimental evaluation is performed on the FIVES retinal vessel segmentation dataset and includes a comparison with a standard U-Net baseline under the same protocol. The results show that anisotropic fuzzy local aggregation supports the extraction of weak and extended vascular fragments and can reduce structural breaks in heterogeneous regions.
Institution: Dubna State University
Language: English
Topic: Retinal Imaging and Analysis
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7202260286