Virus image classification using different texture descriptors

[abstract]

In this work we proposed an ensemble of texture descriptors for virus image classification. Novel variants of texture descriptors, coupled with support vector machines as the classifier, are proposed. The novel variants of texture descriptors include: 1) a quinary coding of different local binary pattern variants, 2) two new approaches based on quinary coding (a selected multithreshold local quinary pattern and a selected multithreshold local quinary configuration pattern), 3) a new approach based on the co-occurrence matrix, and 4) an ensemble of local phase quantization variants with ternary encoding. Our system is compared with and shown to outperform several state of the art texture descriptors. These results are validated on a dataset of 1500 images with 15 classes.

Keywords texture descriptors; feature selection; local phase quantization; local binary patterns; co-occurrence matrix.