A comparative analysis of the Ground clutter suppression on Meteorological images

Local Binary Pattern (LBP) is a gray scale-invariant local operator, it is encoded with each pixel by neighbourhood thresholding which is considered as powerful and simple in the case of real-time implementation. The latter is efficient in characterizing the texture and can be adapted to different problems. Different descriptors have been developed that are based on LBP . In our case, we chose Local Phase Quantization (LPQ), Local Ternary Patterns (LTP), Binarized Statistical Image Features (BSIF), Multiblock Local Binary Pattern (MB −LBP ) to characterize meteorological images and make a comparative analysis between these different methods. Since the objective is to distinguish between precipitation and non-precipitation pixels, we inject the characteristic at the input of a SV M classifier. The approaches were performed on two databases taken from Setif (Algeria) and Bordeaux (France). The experiments show that the best results are obtained by combining LBP to the variance for Setif while the addition of BSIF and LPQ gives the best results when considering the Bordeaux region.