CHILD: A Robust Computationally-Efficient Histogram-based Image Local Descriptor

Designing a robust image local descriptor for the purpose of pattern recognition and classification has been an active area of research. Towards this end, a number of local descriptors based on Weber's law have been proposed recently. Notable among them are Weber Local Descriptor (WLD), Weber Local Binary Pattern (WLBP) and Gabor Weber Local Descriptor (GWLD). Experiments reveal their inability to classify patterns under noisy environments. Our analysis indicates that the components of the WLD: differential excitation and orientation are to be redesigned for robustness and computational efficiency. In this paper, we propose a local Descriptor, called Computationally-Efficient Histogram-based Image Local Descriptor (CHILD), which implements the differential excitation component based on LoG as in WLBP and the orientation component based on fractional order derivatives. The novelty of CHILD stems from the fact that both the LoG and the fractional order derivatives are robust against noise. Further, we have used Wasserstein distance metric to compute the distance between the two histograms and a nearest neighbour classifier which are computationally efficient. We have demonstrated that on the benchmark texture database KTH-TIPS2-a, under both noiseless and noisy conditions, the average classification accuracies of the CHILD consistently outperform the popular local descriptors for texture classification quoted in the literature until 2013.