A Hybrid Guided Filtering and Transform-Based Sparse Representation Framework for Fusion of Multimodal Medical Images

Abstract

Fusion of medical images aims at integrating the harmonizing features of medical images obtained from multimodalities to generate a single image which possess superior visual quality, thus aiding the process of clinical diagnosis in a better way. The challenging task exists when the significant features are extracted simultaneously using the multi-scale transform (MST) methods. To overcome the above-mentioned limitation, fusion framework for multimodal medical images is proposed at two-scale level. Using the two-scale framework, both the structural and texture information from the input medical images are extracted by performing the decomposition of images using a guided filter technique. The base layers are fused in order to preserve the structural details from the images using the integrated DWT and SR pair rule in which the meaningless details are excluded from the source images by constructing the image patch selection-based dictionary, thus enhancing the sparse depiction proficiency of the DWT-decomposed low-frequency layer. A guided filter scheme is applied to combine the detailed layers by enhancing the level of contrast by filtering the noise to the possible level. Finally, the base and detailed parts are combined together to attain the fused image. Experimental results of the proposed framework are assessed objectively which demonstrates the better visual quality by preserving the significant and meaningful features by eliminating the meaningless information.