Boosting local binary pattern with bag-of-filters for content based image retrieval

Local descriptors are mostly extracted over the raw intensity values of the image and requires the analysis of patterns in local pixel neighborhoods. The raw images don't provide much local relationship information required for the local descriptors. To resolve this problem, a new method of local image feature description is proposed in this paper based on the patterns extracted from the multiple filtered images. The whole process is structured as follows: First, the images are filtered with the bag of filters (BoF) and then local binary pattern (LBP) is computed over each filtered image and finally concatenated to find out the single BoF-LBP descriptor. The local information enriches in the filtered images improves the discriminative ability of the final descriptor. Content based image retrieval (CBIR) experiments are performed to observe the effectiveness of the proposed approach and compared with the recent state-of-the-art methods. The experiments are conducted over four benchmark databases having natural and texture images such as Corel-1k, Corel-10k, MIT-VisTex and STex-512S. The experimental results confirm that the introduced approach is able to improve the retrieval performance of the CBIR system.