The analysis of connectivity between parcellated regions of cortex provides insights into the functional architecture of the brain at a systems level. However, there has been less progress in the derivation of functional structures from voxel-wise analyses at finer scales. We propose a novel method, called localized topo-connectivity mapping with singular-value-decomposition-informed filtering (or filtered LTM), to identify and characterize voxel-wise functional structures in the human brain using resting-state fMRI data. Here we describe its mathematical background and provide a proof-of-concept using simulated data that allow an intuitive interpretation of the results of filtered LTM. The algorithm has also been applied to 7T fMRI data as part of the Human Connectome Project to generate group-average LTM images. Functional structures revealed by this approach agree moderately well with anatomical structures identified by T1-weighted images and fractional anisotropy maps derived from diffusion MRI. Moreover, the LTM images also reveal subtle functional variations that are not apparent in the anatomical structures. To assess the performance of LTM images, the subcortical region and occipital white matter were separately parcellated. Statistical tests were performed to demonstrate that the synchronies of fMRI signals in LTM-informed parcellations are significantly larger than those of random parcellations. Overall, the filtered LTM approach can serve as a tool to investigate the functional organization of the brain at the scale of individual voxels as measured in fMRI.