In this chapter we make an extensive study of different state-of-the-art classifiers for building an heterogeneous ensemble for sub-cellular image classification. As features for representing each image we used local ternary patterns. Our aim is to show that it is possible to boost the performance of a stand-alone texture descriptor (here we use the high performance method named local ternary patterns) by an heterogeneous ensemble. First, we compare different classification approaches (different kind of boosting; SVM with various kernels; diverse recent ensemble of decision trees.) in five datasets; then, we show that an heterogeneous ensemble, based on the fusion of different classifiers, performs consistently well across all the tested datasets. The most important result is showing that some very recent approaches and our proposed ensemble outperform also SVM classifier (the well known and widely used LibSVM implementation), even when both kernel selection and the various SVM parameters are carefully tuned. Finally we validated our ensemble also using several datasets from the UCI Repository and other standard pattern classification problems. The Matlab code of the classifiers used in the proposed ensemble is available at bias.csr.unibo.it/nanni/HET.rar. © 2014 Springer-Verlag Berlin Heidelberg.
Nanni, L., Paci, M., Severi, S. (2014). Heterogeneous ensemble of classifiers for sub-cellular image classification based on local ternary patterns. Berlin : Springer Verlag [10.1007/978-3-642-39289-4_6].
Heterogeneous ensemble of classifiers for sub-cellular image classification based on local ternary patterns
NANNI, LORIS;PACI, MICHELANGELO;SEVERI, STEFANO
2014
Abstract
In this chapter we make an extensive study of different state-of-the-art classifiers for building an heterogeneous ensemble for sub-cellular image classification. As features for representing each image we used local ternary patterns. Our aim is to show that it is possible to boost the performance of a stand-alone texture descriptor (here we use the high performance method named local ternary patterns) by an heterogeneous ensemble. First, we compare different classification approaches (different kind of boosting; SVM with various kernels; diverse recent ensemble of decision trees.) in five datasets; then, we show that an heterogeneous ensemble, based on the fusion of different classifiers, performs consistently well across all the tested datasets. The most important result is showing that some very recent approaches and our proposed ensemble outperform also SVM classifier (the well known and widely used LibSVM implementation), even when both kernel selection and the various SVM parameters are carefully tuned. Finally we validated our ensemble also using several datasets from the UCI Repository and other standard pattern classification problems. The Matlab code of the classifiers used in the proposed ensemble is available at bias.csr.unibo.it/nanni/HET.rar. © 2014 Springer-Verlag Berlin Heidelberg.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.