A measurement of the top quark mass is presented using top quark candidates reconstructed from the jets in all-hadronic top-antitop decays. The analysis makes use of the full ATLAS dataset collected over the 2012 period at a centre-of-mass energy of 8 TeV, with a total integrated luminosity of 20.6 inverse femtobarns. A five-jet trigger, together with an offline cut requiring five central jets with a minimum transverse momentum of 60 GeV, was used to pre-select candidate signal events. A series of selection cuts are subsequently employed to increase the signal fraction in data events, motivated by both data and simulation. These cuts, as well the analytic chi-squared reconstruction algorithm to designate which jets are used to reconstruct the events, aim to select those events and candidate top quarks deemed most consistent with the signal topology. The most discriminating cut involves the output of a b-tagging algorithm employed to identify jets initiated by a bottom-type quark.

The measurement is made using a one-dimensional template method, in which a series of distributions are produced for an observable sensitive to the top quark mass by using simulated signal samples with varying input mass values. The ratio of the invariant masses of candidate top quark - W boson pairs is selected as the top quark mass-sensitive observable, as it is less susceptible to uncertainties arising from the jet energy scale, thereby reducing the corresponding systematic uncertainty. The contributions from backgrounds, dominated by QCD multi-jet events, are estimated using a mix of simulation and data-driven methods.

The final measurement employs a binned chi-squared minimization procedure which takes into account uncertainties in the bin contents from the measured data as well as uncertainties in the parameterization for both signal and background shapes.

The final measured top quark mass value is found to be 174.29 +/- 0.52 (stat) +/- 1.03 (syst) GeV, consistent both with recent LHC measurements and the current world average. The fraction of background events within the range of the final distribution, measured simultaneously, is found to be 0.517 +/- 0.015 (stat) +/- 0.075 (syst).