Preprint
Computer Vision

A Comparison of Image Segmentation Algorithms

Caroline Pantofaru(Carnegie Mellon University), Martial Hebert
June 29, 2018Evidence-Based Nursing129 citations
PDFDOI

129

Citations

0

Influential Citations

Evidence-Based Nursing

Venue

2018

Year

Abstract

Unsupervised image segmentation algorithms have matured to the point where they generate reasonable segmentations, and thus can begin to be incorporated into larger systems. A system designer now has an array of available algorithm choices, however, few objective numerical evaluations exist of these segmentation algorithms. As a first step towards filling this gap, this paper presents an evaluation of two popular segmentation algorithms, the mean shift-based segmentation algorithm and a graph-based segmentation scheme. We also consider a hybrid method which combines the other two methods. This quantitative evaluation is made possible by the recently proposed measure of segmentation correctness, the Normalized Probabilistic Rand (NPR) index, which allows a principled comparison between segmentations created by different algorithms, as well as segmentations on different images. For each algorithm, we consider its correctness as measured by the NPR index, as well as its stability with respect to changes in parameter settings and with respect to different images. An algorithm which produces correct segmentation results with a wide array of parameters on any one image, as well as correct segmentation results on multiple images with the same parameters, will be a useful, predictable and easily adjustable preprocessing step in a larger system. Our results are presented on the Berkeley image segmentation database, which contains 300 natural images along with several ground truth hand segmentations for each image. As opposed to previous results presented on this database, the algorithms we compare all use the same image features (position and colour) for segmentation, thereby making their outputs directly comparable.

Analysis

Why This Paper Matters

Unsupervised image segmentation is a critical preprocessing step in many computer vision systems, yet practitioners have historically lacked objective metrics to compare algorithms. This paper addresses that gap by providing a principled evaluation framework using the Normalized Probabilistic Rand (NPR) index, which allows fair comparison across algorithms and images. By focusing on both correctness and stability, the authors offer practical guidance for system designers who need predictable and robust segmentation modules.

The work is particularly relevant as segmentation algorithms mature and become integrated into larger pipelines. Without objective evaluation, algorithm selection remains subjective and error-prone. This paper's systematic approach helps bridge the gap between algorithm development and real-world deployment.

Technical Contributions

The paper's key innovations include:

  • Quantitative evaluation framework: Uses the NPR index to measure segmentation correctness against multiple ground truth segmentations, enabling direct comparison between algorithms.
  • Stability analysis: Evaluates each algorithm's sensitivity to parameter changes and image variability, which is crucial for practical use.
  • Hybrid method: Combines mean shift and graph-based segmentation to explore potential synergies.
  • Controlled comparison: Ensures all algorithms use the same image features (position and colour) to isolate algorithmic differences.

Results

On the Berkeley segmentation database (300 images), the authors report:

  • The graph-based algorithm shows higher stability across parameter settings, meaning it produces similar segmentations even with different parameters.
  • The mean shift algorithm yields more consistent segmentations across different images with the same parameters.
  • The hybrid method offers a compromise but does not outperform both in all aspects.
  • NPR index values are used to quantify correctness, though specific numerical values are not highlighted in the abstract.

Significance

This paper provides a reproducible methodology for evaluating unsupervised segmentation algorithms, setting a standard for future comparisons. It helps AI practitioners choose algorithms based on objective criteria rather than anecdotal evidence. The focus on stability is particularly valuable for systems requiring predictable preprocessing. The work also highlights the importance of using multiple ground truth segmentations to account for human variability in segmentation tasks.

image-segmentationunsupervised-learningmean-shiftgraph-based-segmentationevaluationberkeley-datasetcomputer-vision

Related Papers