Volumetric congruent local binary patterns for 3D neurological image retrieval

The high thought put and increasingly accumulated data size of the 3D neuroimaging datasets have posed great challenges for neuroimaging data retrieval. To efficiently manage such large datasets, we proposed a volumetric congruent local binary pattern (vcLBP) algorithm for 3D neurological image retrieval. The vcLBP-based feature descriptor could describe the volumetric imaging data with higher robustness and meanwhile effectively compress the feature space by using the unique rotation, reflection and translation invariant patterns. We evaluated the proposed vcLBP algorithm using 132 sets of 3D positron emission tomography (PET) brain imaging data and the preliminary results suggested that our approach could effectively reduce the feature dimensions while achieving better results than other 3D feature descriptors. This vcLBP algorithm has a potential to be widely used in many other applications, such as image classification, content analysis, and data mining.