Circle-Based Object Counting Using Image Analysis

• Altair Felipe Peiter UFSM
• Bernardo Paul Lorenzoni Ávila UFSM
• Leonardo Ramos Emmendörfer UFSM
• Lucas Both Steinmetz Ribeiro UFSM

Resumo

Accurate estimation of object counts in binary images with spatially clustered or overlapping instances remains a challenging problem in fields such as microscopy, microfluidics, and industrial inspection. Classical segmentation techniques, including contour-based and watershed algorithms, often perform suboptimally in scenarios characterized by occlusion, high object density, or morphological variability. In this work, we present an unsupervised, training-free method for object counting based on stochastic placement of fixed-radius circles within segmented regions, followed by iterative merging using geometric criteria. Experimental results on synthetic datasets with varying degrees of object overlap indicate that the proposed approach provides robust and reliable estimates, frequently outperforming standard baselines, especially in cases of moderate to high agglutination. Qualitative analysis further demonstrates its ability to resolve individual objects within complex clusters while maintaining computational efficiency. These findings highlight the practical potential of the method for unsupervised object counting in challenging imaging scenarios.

Palavras-chave: Image segmentation, Microscopy, Merging, Clustering algorithms, Watersheds, Inspection, Stability analysis, Computational efficiency, Iterative methods, Synthetic data