Fast Adaptive Entropy-driven Multilevel Image Segmentation using Look-up Tables on GPU Platforms

Georgi Petrov
Department of Telecommunications,
New Bulgarian University,
Sofia (Bulgaria)

https://doi.org/10.53656/math2026-1-2-fae

Abstract. This paper proposes a fast, adaptive method for multilevel image segmentation based on Kapur’s entropy. Implemented on CPU and GPU, the method achieves up to 40× acceleration using CUDA and memory optimization. Segmentation quality is preserved while enabling real-time processing of large image batches. The experiments were carried out using the university’s high-performance computing (HPC) mini-cluster, demonstrating its role as an educational and research platform. The approach is scalable and suited for scientific and educational use. It builds upon earlier work on multidimensional histogram analysis, applying entropy-driven modeling to image segmentation.
Keywords: entropy, multilevel thresholding, GPU, LUT

>> Download the article as a PDF file <<

Свързани статии:

The Problem of Multicultural Component of Language Education In Memory of Assoc. Prof. Olga Todorova Diagnosis as a Tool for Monitoring the Effectiveness of Addiction Prevention in Adolescents With a Mission for National and Cultural Individuality (On the Occasion of 135 Years since the Opening of the First Higher Pedagogical Course in Bulgaria (1888)