Anastylosis of Frescos – A Web Service in an HPC Environment

Abstract

The parallelization in a High Performance Computing (HPC) environment of a method, algorithm and software called RINCCAS (Rotation-Invariant NCC for 2D Color Matching of Arbitrary Shaped Fragments of a Fresco) is presented. RINCCAS is designed for virtual restoration of frescoes from their ruins, a known problem of national and world heritage conservation. The method was developed at IICT-BAS for participation in the DAFNE computer competition, Italy, 2019. Prepared in the MATLAB environment, RINCCAS uses a classic NCC approach for positioning of the fragments (rectangular coordinates and accidental rotation angle). It extends NCC to color input images (frescoes and fragments) and to arbitrary shapes of the fragments. RINCCAS is a cubic complexity method. To reduce the execution time, we use parallelization through independent subtasks whose optimal distribution in consecutive sessions we call sequential concatenation within one or more HPC nodes. A description and comparative analysis of the experiments with Avitohol-HPC at IICT-BAS are presented. Upper limits are set for the amount of input data relative to the amount of available RAM, not meeting which significantly slows performance due to system memory swapping. An improvement called parallel concatenation is proposed to significantly alleviate memory constraints using the same theoretical formulation. In conclusion, an invitation is extended to the recently opened website of the RINCCAS-HPC service for Anastylosis of Frescos’ enthusiasts and specialists.