Labeling and marking industrial manufactured objects gets increasingly important nowadays because of novel material properties and plagiarism. As part of the Collaborative Research Center 653 which investigates micro-structured metallic surfaces for inherent mechanical data storage, we research into a stable and reliable optical readout of the written data. Since this comprises a qualitative surface reconstruction, we use directed illumination to make the micro structures visible. Then we apply a spectral analysis to obtain image partitioning and perform signal tracking enhanced by a customized Hidden Markov Model. In this paper, we derive the algorithms used and demonstrate reading data from a surface with 1.6 kbit/cm2 from a micro-structured groove which varies by only 3 ÎŒm in depth (thus a âscratchâ). We demonstrate the systemâs robustness with experiments with real and artificially-rendered surfaces.