The ATLAS Pixel detector is currently measuring particle positions at 8 TeV proton-proton collisions at the LHC. In the dense environment of jets with high transverse momenta produced in these events the separation between particles becomes small, such that their respective charge deposited are reconstructed as single clusters. A Neural Network (NN)-based clustering algorithm has been developed to identify such merged clusters. By using all cluster information, the NN is ideal to estimate the particle multiplicity and for each of the estimated number of particles, the position with its uncertainty. As a result of the NN reconstruction, the number of hits shared by several tracks is strongly reduced. Furthermore, the impact parameter improves by about 15% which indicates boosted prospects for physics analysis.