ROMJIST Volume 29, No. 2, 2026, pp. 103-114, DOI: 10.59277/ROMJIST.2026.2.01
Mehmet FIDAN, Semih ERGIN, Mehmet KOC, Mehmet Bilginer GULMEZOGLU, Omer Nezih GEREK, Atalay BARKANA Ratio-Preserving Ellipse-Fit Alignment for Robust 2-D Shape Recognition
ABSTRACT: A lightweight yet powerful pipeline for 2-D shape recognition is presented, preserving each shape’s intrinsic aspect ratio while aligning it to every reference candidate. The approach first fits ellipses to query and reference shapes, then rescales the query’s minor axis so that its original major-to-minor ratio is kept unchanged after mapping to the reference ellipse. A single affine matrix obtained from the updated vertices compensates for translation, rotation, and global scale in one step. This work introduces a novel ratio-preserving ellipse-fit alignment (RPEF) technique that significantly enhances robustness in 2-D shape recognition, particularly under varying orientations and scales. Unlike traditional approaches such as the Randomized Hough Transform or HyperLS, which struggle with noisy data or iterative complexities, RPEF ensures numerical stability and efficiency by leveraging a geometrically informed prior. The approach’s key contribution lies in its ability to maintain shape integrity without requiring extensive training data, making it highly adaptable for applications with limited labeled datasets, such as medical image analysis or industrial inspection. Additionally, the pipeline’s one-step affine transformation simplifies the alignment process, outperforming existing approaches like LShape descriptors by achieving superior accuracy metrics, 100% on Kimia-99 and 85.5% on TARI-1000, demonstrating its practical utility and effectiveness in real-world scenarios.KEYWORDS: 2-D shape recognition; affine transformation; ellipse fitting; ratio-preserving scaling; similarity metrics.Read full text (pdf)