Biomedical image indexing and retrieval based on new efficient hybrid approach using directional decomposition and a novel local directional frequency encoded pattern: a post feature descriptor

Abstract

In this paper, a new efficient hybrid approach based on directional decomposition and post local feature extraction is proposed for biomedical image indexing and retrieval. Initially, triplet half-band filter bank (THFB) is modified and used in directional filter bank (DFB) for directional decomposition of images. DFB decomposes image into directional frequency sub-bands. To acquire local information in each directional sub-band of images, a novel local directional frequency encoded pattern (LDFEP) feature descriptor is proposed as post feature descriptor. The LDFEP is based on establishing the relationship between 00, 450, 900, and 1350 directional frequency components. The deliberation of directional as well as local information composes hybrid approach which is more efficient than the existing descriptors for biomedical image retrieval. Manhattan distance is selected to compute analogy between the query feature vector and the feature vector of images from the database. The efficacy of the proposed approach in terms of precision and recall has been evaluated by conducting the experiments on three well-known biomedical databases: Open access series of imaging studies (OASIS)-MRI, EXACT 09-CT and NEMA-CT. The experimental results confirmed the superiority of the proposed approach in comparison with the state-of-the-art feature descriptors for biomedical image retrieval.