Dense-Label-Map-Free SLIC Segmentation via Bit-Mask Cluster Encoding
IEEE Signal Processing Letters · 2026 · 33 · 2310-2314
DOI: 10.1109/lsp.2026.3699336 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Yuan Li und 4 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben.
Abstract (Originalsprache)
Superpixels are widely used as a preprocessing step to reduce the computational cost of computer vision systems. Simple Linear Iterative Clustering (SLIC) is one of the most popular methods due to its simplicity and effectiveness. However, conventional SLIC requires storing a full image size label map throughout the iterative assignment and update process, leading to high memory usage and limited computational efficiency. This letter presents a SLIC implementation that replaces the dense image-sized label map with bit-mask cluster encoding. Instead of storing per-pixel labels for the entire image, each superpixel cluster stores its center parameters and a local bit mask to record pixel membership. A displacement-based early termination mechanism is further incorporated as a practical engineering optimization to stop iteration when the change between the current and previous displacement becomes smaller than a fixed fraction of that in the current iteration. The proposed method enables a highly efficient realization of SLIC, reducing memory consumption, computation, and runtime while preserving segmentation quality. Experiments on BSDS500 show an average processing time of 3.149 ms, and the Kintex-7 FPGA implementation uses only 3.58 % LUTs and 26.74 % BRAM.
Zitieren
APA
Li, Y., Wang, W., Liu, Y., Wang, Z., & Shi, Y. (2026). Dense-Label-Map-Free SLIC Segmentation via Bit-Mask Cluster Encoding. IEEE Signal Processing Letters, 33, 2310-2314. https://doi.org/10.1109/lsp.2026.3699336
Vancouver
Li Y, Wang W, Liu Y, Wang Z, Shi Y. Dense-Label-Map-Free SLIC Segmentation via Bit-Mask Cluster Encoding. IEEE Signal Processing Letters. 2026;33:2310-2314. doi:10.1109/lsp.2026.3699336
BibTeX
@article{Li2026,
author = {Li, Yuan and Wang, Wei and Liu, Yanchao and Wang, Zihang and Shi, Youhua},
title = {Dense-Label-Map-Free SLIC Segmentation via Bit-Mask Cluster Encoding},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {2310--2314},
doi = {10.1109/lsp.2026.3699336}
}
EUVIMED