Recent research efforts allowed combining image acquisition with limit analysis to investigate failure mechanisms and limit loads in irregular historical masonry. These studies were however so far predominantly restricted to 2D analysis [1]. The present contribution proposes a methodology for the extraction of 3D computational geometries of blocky structures from images to be subsequently used in the evaluation of the corresponding failure mechanisms and limit loads, including complex load cases such as out- of-plane loading. A kinematic limit analysis procedure is considered, in which blocks are assumed infinitely rigid and strong, restricting power dissipation – and thus failure – to joints only. These joints are represented as zero-thickness interfaces located on the so-called medial surface of the block arrangement. Given block positions, the computational geometry construction follows a multi-step approach: (i) blocks are labeled, and their centroids are extracted and used in a constrained centroidal Voronoï tessellation. (ii) Global distance fields on a regular grid are computed. Following the level-set approach proposed in [2], the medial surface is computed using level- set functions constructed from the distance maps to the first, second and third nearest inclusions. Multiplicities – which determine the underlying geometric entity of a grid point – are determined. In a regular 3D setting, multiplicity 2, 3 and ≥4 denote faces, edges and vertices (quadruple points) respectively. (iv) Voronoï cell vertices are displaced to their correct locations – corresponding to quadruple points – based on the level-set functions and gradients associated with block triads. (v) edge points are introduced along segments connecting quadruple points and repositioned using the same Voronoï-like functions and gradients. (vi) Finally, the surface is reconstructed by placing random points on the medial surface contours and mapping them to nearby multiplicity 1 points, allowing triangulation with standard methods. The obtained surfaces serve as joint representations in block limit analysis procedures.