Novel local bit-plane dissimilarity pattern for computed tomography image retrieval

摘要

The performance of any image retrieval system relies over the image descriptor used for matching. The recently proposed descriptors mainly suffer due to the ‘less discriminative power’ and ‘curse of dimensionality’. A novel image descriptor using local bit-plane dissimilarity pattern (LBDISP) is introduced. The descriptor is computed by finding the dissimilarity map between the centre pixel and its neighbours over each bit-plane. The relationship between the centre pixel and dissimilarity map is then encoded to form the LBDISP descriptor. The dimension of the descriptor depends upon the number of neighbours only. A retrieval experiment over NEMA-CT database confirms the superiority and efficiency of the LBDISP descriptor.