Resolving Overlapping Convex Objects in Silhouette Images by Concavity Analysis and Gaussian Process

Sahar Zafari, Mariia Murashkina, Tuomas Eerola, Jouni Sampo, Heikki Kälviäinen, Heikki Haario

Abstract

Segmentation of overlapping convex objects has various applications, for example, in nanoparticles and cell imaging. Often the segmentation method has to rely purely on edges between the background and foreground, making the analyzed images essentially silhouette images. Therefore, to segment the objects, the method needs to be able to resolve the overlaps between multiple objects by utilizing prior information about the shape of the objects. This paper introduces a novel method for segmentation of clustered partially overlapping convex objects in silhouette images. The proposed method involves three main steps: pre-processing, contour evidence extraction, and contour estimation. Contour evidence extraction starts by recovering contour segments from a binarized image by detecting concave points. After this, the contour segments which belong to the same objects are grouped. Finally, the full contours of the objects are estimated by a Gaussian process regression method. The experiments on a challenging dataset consisting of nanoparticles demonstrate that the proposed method outperforms three current state-of-the-art approaches.

Introduction

In fields such as nanotechnology and biomedicine, researchers frequently encounter images of "cluttered" objects, such as clusters of nanoparticles or cells. A significant challenge arises when these images are silhouettes, where internal edges between overlapping objects are invisible. Traditional methods like the Watershed transform often over-segment, while rigid models like ellipse fitting fail for non-ideal shapes. Deep learning approaches, though powerful, require extensive manual annotation often unavailable for niche scientific data. This paper addresses the need for a geometry-based approach that can resolve overlaps using only visible boundary information and the assumption of object convexity.

A cluster of overlapping nanoparticles illustrating the challenge of silhouette segmentation.

Methodology

The proposed pipeline consists of three modular stages:

  1. Pre-processing: Images are binarized using Otsu’s method, and edges are extracted via a Canny detector to create a silhouette map.
  2. Contour Evidence Extraction:
    • Concave Point Detection: The algorithm identifies "pinch points" where objects meet using a parameter-free adaptation of the Prasad+Zhang method, which automatically calculates thresholds based on digital line characteristics.
  3. Segment Grouping: Boundary segments are grouped into individual objects by solving a combinatorial optimization problem via a Branch and Bound (BB) algorithm.
  4. Contour Estimation: The primary contribution is the use of Gaussian Process Regression in Polar Form (GP-PF). By transforming segments into polar coordinates ($r, \theta$), the contour is treated as a periodic function. The Gaussian Process "fills in" occluded parts, allowing for flexible, non-parametric shape reconstruction that respects the "mostly convex" nature of the objects without being restricted to rigid templates like ellipses.
The proposed three-step pipeline: Pre-processing, Contour Evidence Extraction (Concave Point Detection and Segment Grouping), and Contour Estimation using Gaussian Processes.

Experimental Results

The method was evaluated on synthetic datasets of overlapping ellipses and a real-world nanoparticle dataset containing approximately 2,200 manually annotated particles.

  • Performance: On the nanoparticle dataset, the proposed method achieved a True Positive Rate (TPR) of 87% and an Accuracy (ACC) of 74%, significantly outperforming the next best method (SCC at 62%).
  • Shape Precision: Qualitative analysis indicates that GP-PF produces more realistic boundaries for occluded regions compared to rigid ellipse fitting or B-splines, which often misestimate hidden areas.
  • Robustness: The automated thresholding for concave point detection makes the method portable across different imaging conditions without manual tuning.

Conclusion

This research presents a robust framework for segmenting overlapping convex objects in silhouette images. By leveraging Gaussian Processes in polar coordinates, the method provides a flexible alternative to rigid shape models. While currently constrained by the convexity assumption and the computational cost of the Branch and Bound algorithm, it offers a significant advancement for automated analysis in microscopy and materials science. Future work will explore extending the model to non-convex shapes and optimizing computational efficiency.

Summary updated

Abstract

Segmentation of overlapping convex objects has various applications, for example, in nanoparticles and cell imaging. Often the segmentation method has to rely purely on edges between the background and foreground making the analyzed images essentially silhouette images. Therefore, to segment the objects, the method needs to be able to resolve the overlaps between multiple objects by utilizing prior information about the shape of the objects. This paper introduces a novel method for segmentation of clustered partially overlapping convex objects in silhouette images. The proposed method involves three main steps: pre-processing, contour evidence extraction, and contour estimation. Contour evidence extraction starts by recovering contour segments from a binarized image by detecting concave points. After this, the contour segments which belong to the same objects are grouped. The grouping is formulated as a combinatorial optimization problem and solved using the branch and bound algorithm. Finally, the full contours of the objects are estimated by a Gaussian process regression method. The experiments on a challenging dataset consisting of nanoparticles demonstrate that the proposed method outperforms three current state-of-art approaches in overlapping convex objects segmentation. The method relies only on edge information and can be applied to any segmentation problems where the objects are partially overlapping and have a convex shape.
BibTeX
@article{zafari2020resolving,
  author = {Sahar Zafari and Mariia Murashkina and Tuomas Eerola and Jouni Sampo and Heikki Kälviäinen and Heikki Haario},
  title = {{Resolving Overlapping Convex Objects in Silhouette Images by Concavity Analysis and Gaussian Process}},
  journal = {J. Vis. Commun. Image Represent.},
  year = {2020},
  url = {https://openreview.net/forum?id=QV8ikZrx3u},
}
Figure