Adequately hierarchical patterns based on pairwise regions

Thanh Tuan Nguyen, Thanh Phuong Nguyen, Frédéric Bouchara

Abstract: We propose a novel operator, called CHILOP, for capturing the hierarchical relationships of two adjacent supporting regions in a completed local encoding context. It can be seen as a generalization of typical CLBP, one of the most popular local operators. CHILOP is then considered in a multi-scale approach to forcefully capture multi-hierarchical patterns with more robustness. Moreover, multi-hierarchical Gaussian-filtered CHILOP properties are taken into account for a more discriminative descriptor inspired by filter-bank approaches. A comprehensive evaluation on different benchmark datasets has proven the benefit of our proposed method.