Dynamic Texture Representation Based on Hierarchical Local Patterns

Thanh Tuan Nguyen (1) , Thanh Phuong Nguyen (1) , Frédéric Bouchara (1)

1 SIIM - Signal et Image

A novel effective operator, named HIerarchical LOcal Pattern (HILOP), is proposed to efficiently exploit relationships of local neighbors at each adjacent pairwise of different regional hierarchies located surrounding a center pixel of a texture image. Instead of thresh-olding by the value of central pixel, the gray-scale of each local neighbor in a hierarchical area is compared to that of all of neighbors in the remain region. In order to capture shape and motion cues for dynamic texture (DT) representation, HILOP is taken into account for investigating hierarchical relationships in plane images of a DT sequence. The obtained histograms are then concatenated to form a robust descriptor with high performance for DT classification task. Experiments on various benchmark datasets (i.e., UCLA, DynTex, DynTex++) have validated the interest of our proposal.

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]; Mathématique discrète [cs.DM]

Thanh Tuan Nguyen, Thanh Phuong Nguyen, Frédéric Bouchara. Dynamic Texture Representation Based on Hierarchical Local Patterns. Advanced Concepts for Intelligent Vision Systems, Feb 2020, Auckland, New Zealand. ⟨hal-02357082⟩