Analyzing features by SWLDA for the classification of HEp-2 cell images using GMM

•Fading of a script alone does not foster domain-general strategy knowledge.•Performance of the strategy declines during the fading of a script.•Monitoring by a peer keeps performance of the strategy up during script fading.•Performance of a strategy after fading fosters domain-general strategy know...

Full description

Saved in:
Bibliographic Details
Published in:Pattern recognition letters 2016-10, Vol.82 (P1), p.44-55
Main Authors: Sarrafzadeh, Omid, Rabbani, Hossein, Mehri Dehnavi, Alireza, Talebi, Ardeshir
Format: Article
Language:English
Subjects:
Automatic system
Automation
Cells
Discriminant analysis
Epithelial cells
Feature extraction
Fluorescence
Gaussian Mixture Model
Image classification
Image processing systems
Immunofluorescence
Indirect immunofluorescence
Normal distribution
Pattern recognition
Step-Wise Linear Discriminant Analysis
Wavelet analysis
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•Fading of a script alone does not foster domain-general strategy knowledge.•Performance of the strategy declines during the fading of a script.•Monitoring by a peer keeps performance of the strategy up during script fading.•Performance of a strategy after fading fosters domain-general strategy knowledge.•Fading and monitoring by a peer combined foster domain-general strategy knowledge. In this paper, a system is introduced for automatic classification of Human Epithelial cells type 2 Patterns (HEp-2) in indirect immunofluorescence imaging. HEp-2 cell classification was performed using Step-Wise Linear Discriminant Analysis (SWLDA) and Gaussian Mixture Model (GMM). Images were first normalized. Then, binary, intensity, statistical, spectral, wavelet-based, Haralick, CLBP and Gabor features were extracted from the normalized images. The best features were then selected using SWLDA, and the GMM framework was used for classification. Two protocols were examined considering all data and divided data (into intermediate and positive groups). In the first protocol all data are modeled with one GMM and in the second protocol two GMM models are designed for intermediate and positive data. The methods were applied on the ICPR2012 and ICIP2013 datasets. For the ICPR2012 dataset, a third protocol was also proposed based on the results of the second protocol. The classification was evaluated using standard metrics. The comparative results show that our method outperformed previous works for the ICPR2012 dataset and intermediate for the ICIP2013 dataset. [Display omitted]
ISSN:0167-8655
1872-7344
DOI:10.1016/j.patrec.2016.03.023