Abstract
This document presents an improved global descriptor, the Robust Visual Descriptor (RVD). Our proposed global descriptor results in an improvement of the overall performance of the Test Model TM6 in both pairwise matching and retrieval, showing exceptional performance improvements in the non-planar 3D objects experiments, where the TM currently under-performs. Furthermore, our proposed global descriptor reduces the memory required in the encoder by ~25%. Finally, the proposed method entails a very simple control mechanism for the global descriptor matching process, using just a single threshold to control the balance between True Positive (TPR) and False Positive (FPR) rates in pairwise matching.