Related Experiment Video
Updated: Jun 10, 2026
10:39
A Label-Free Segmentation Approach for Intravital Imaging of Mammary Tumor Microenvironment
Published on: May 24, 2022
An extension of the standard mixture model for image segmentation
Thanh Minh Nguyen1, Q M Jonathan Wu, Siddhant Ahuja
1Department of Electrical and Computer Engineering, University of Windsor, Windsor, ON N9B-3P4, Canada. nguyen1j@uwindsor.ca
IEEE Transactions on Neural Networks
|July 21, 2010
Summary
This study introduces an improved Gaussian Mixture Model (GMM) for image segmentation, enhancing noise robustness by incorporating spatial pixel relationships. The novel method offers an effective and efficient alternative to existing Gaussian Mixture Models and Markov Random Field models.
Area of Science:
- Computer Vision
- Image Processing
- Machine Learning
Background:
- Standard Gaussian Mixture Models (GMM) for image segmentation treat pixels as independent, leading to noise sensitivity.
- Markov Random Field (MRF) models account for spatial dependencies but are computationally intensive and parameter-heavy.
Purpose of the Study:
- To develop an extended GMM for image segmentation that incorporates spatial pixel relationships.
- To create a model that is less sensitive to noise than standard GMM and more computationally efficient than MRF models.
Main Methods:
- A novel approach integrates spatial relationships between neighboring pixels into the standard GMM framework.
- A new parameter estimation method minimizes the upper bound on data negative log-likelihood using gradient-based optimization.
Main Results:
- The proposed model demonstrates robustness and accuracy in segmenting noisy synthetic and real-world grayscale images.
- Experimental results show superior performance compared to standard GMM and MRF-based methods.
Conclusions:
- The proposed GMM extension effectively incorporates spatial information for improved image segmentation.
- The model offers a computationally efficient and parameter-light alternative for noise-robust image segmentation.
