Modified Binary Robust Independent Elementary Features for fast Earth Observation Data Analysis

Kurzfassung

In this paper a new image descriptor based on Binary Robust Independent Elementary Features (BRIEF) is presented. The proposed algorithm uses the mean intensity of predefined regions within the analyzed patch to compute the common binary tests of BRIEF. Therefore our method can be expended for patch classification instead of patch or image matching. This approach is speeding up the descriptor computation for large patches, compared with classical methods as Gabor and Weber Local Descriptors (WLD) but it doesn’t improve the classification performances. As similarity measure we used the Hamming distance. The goal is to demonstrate the usability of this new binary descriptor in Earth Observation image analysis by comparison with standard texture descriptors as Gabor and WLD.