Deterministic annealing EM and its application in natural image segmentation

Authors:
Jonghyun Park;Wanhyun Cho;Soonyoung Park
Affiliations:
Department of Computer Science, Chonbuk National University, S. Korea;Department of Statistics, Chonnam National University, S. Korea;Department of Electronic Engineering, Mokpo National University, S. Korea
Venue:
CIS'04 Proceedings of the First international conference on Computational and Information Science
Year:
2004

Citing 5
Cited 3

Deterministic annealing EM algorithm

Neural Networks
Unsupervised Segmentation of Color-Texture Regions in Images and Video

IEEE Transactions on Pattern Analysis and Machine Intelligence
Unsupervised Learning of Finite Mixture Models

IEEE Transactions on Pattern Analysis and Machine Intelligence
Pairwise Data Clustering by Deterministic Annealing

IEEE Transactions on Pattern Analysis and Machine Intelligence
Automatic detection and recognition of signs from natural scenes

IEEE Transactions on Image Processing

Color image segmentation using adaptive mean shift and statistical model-based methods

Computers & Mathematics with Applications
Noise removal and restoration using voting-based analysis and image segmentation based on statistical models

EMMCVPR'07 Proceedings of the 6th international conference on Energy minimization methods in computer vision and pattern recognition
A novel method for image retrieval using relevance feedback and unsupervised clustering

COMPUTE '11 Proceedings of the Fourth Annual ACM Bangalore Conference

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this paper, we present a color image segmentation algorithm based on a finite mixture model and examine its application to natural scene segmentation. Gaussian mixture model (GMM) is first adopted to represent the statistical distribution of multi-colored objects. Then a deterministic annealing Expectation Maximization (DAEM) formula is used to estimate the parameters of the GMM. The experimental results show that the proposed DAEM can avoid the initialization problem unlike the standard EM algorithm during the maximum likelihood (ML) parameter estimation and natural scenes containing texts are segmented more efficiently than the existing EM technique.