•Algorithms in computer vision are customized and complimented for concrete CT.•3D information of 1442 aggregates is extracted from the bulk concrete.•3D percolation appears in the aggregates while they are isolated in 2D.•The heterogeneous modelling with individual aggregate control is 3D printable.•The presented work enables 3D concrete modelling with realistic aggregates input. A computational sequence to extract individual 3D aggregate from computed tomography (CT) scans of bulk concrete is presented in this paper. The technique adopted a combination of computer vision algorithms customized for concrete composite along with the originally proposed algorithm. The computational technique efficiently establishes a database of 1442 aggregates with individual 3D printable volume and surface. The presented work can be used to perform heterogeneous modelling with realistic aggregates input and distributive analysis of a specific constituent of interest. The necessity of customizing the computer vision in the application of concrete composite is discussed with evidenced digital damage in the non-destructive measurement. A comparison with available reconstruction methods with feasibility is performed to demonstrate that characterizing concrete composite with detailed information of each constituent can provide a more realistic representation of the composite. The vision of digital concrete with realistic aggregates input is further demonstrated with the example application.