Image classification with binary gradient contours

In this work we present a new family of computationally simple texture descriptors, referred to as binary gradient contours (BGC). The BGC methodology relies on computing a set of eight binary gradients between pairs of pixels all along a closed path around the central pixel of a 3×3 grayscale image patch. We developed three different versions of BGC features, namely single-loop, double-loop and triple-loop. To quantitatively assess the effectiveness of the proposed approach we performed an ensemble of texture classification experiments over 10 different datasets. The obtained results make it apparent that the single-loop version is the best performer of the BGC family. Experiments also show that the single-loop BGC texture operator outperforms the well-known LBP. Statistical significance of the achieved accuracy improvement has been demonstrated through the Wilkoxon signed rank test.

► We present a novel family of powerful and computationally simple texture descriptors: the binary gradient contours (BGC).
► We also present a general framework for texture description based on grayscale pattern mapping.
► We developed three different versions of BGC features, namely single-loop, double-loop and triple-loop.
► We found that single-loop BGC texture descriptor (BGC1) is the best performer of the BGC family.
► Experiments show that the BGC1 texture operator is more accurate than the well-known LBP.