In this paper we propose a new feature extractor technique for pattern classification that is based on the calculation of texture descriptors. Starting from the standard feature vector representation, we rearrange the patterns as matrices and then apply such standard texture descriptor techniques as local binary patterns, local ternary patterns, and Coiflet wavelets. In our classification experiments using several well-known benchmark datasets, support vector machines are used both for the vector-based descriptors and the texture descriptors. Using our new feature extractor technique, the feature vector is arranged as a matrix by random assignment. For each pattern, 50 different random assignments are performed, and then the classification results are combined using the mean rule. We believe that our novel technique introduces a new source of information. Our experiments show that the texture descriptors along with the vector-based descriptors can be combined to improve overall classifier performance. In our experimental results the performance obtained by our extraction technique outperformed that obtained by support vector machines trained using standard vector-based descriptors.
L. Nanni, S. Brahnam, A. Lumini (2011). Texture descriptors for generic pattern classification problems. EXPERT SYSTEMS WITH APPLICATIONS, 38(8), 9340-9345 [10.1016/j.eswa.2011.01.123].
Texture descriptors for generic pattern classification problems
NANNI, LORIS;LUMINI, ALESSANDRA
2011
Abstract
In this paper we propose a new feature extractor technique for pattern classification that is based on the calculation of texture descriptors. Starting from the standard feature vector representation, we rearrange the patterns as matrices and then apply such standard texture descriptor techniques as local binary patterns, local ternary patterns, and Coiflet wavelets. In our classification experiments using several well-known benchmark datasets, support vector machines are used both for the vector-based descriptors and the texture descriptors. Using our new feature extractor technique, the feature vector is arranged as a matrix by random assignment. For each pattern, 50 different random assignments are performed, and then the classification results are combined using the mean rule. We believe that our novel technique introduces a new source of information. Our experiments show that the texture descriptors along with the vector-based descriptors can be combined to improve overall classifier performance. In our experimental results the performance obtained by our extraction technique outperformed that obtained by support vector machines trained using standard vector-based descriptors.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.