Despite recent advancements in powder bed fusion additive manufacturing, in-process monitoring and quality assurance remain insufficient for wide applications in production. One obstacle is the interpretation and evaluation of the intense monitoring data. The objective of this study is to discuss a data processing infrastructure for coaxial sensing toward consistent and repetitive additive manufacturing. The infrastructure is developed based on an image database collected from an in-house-developed powder bed fusion platform. This work first addresses salient issues in the raw data (e.g., noises, inconsistent illumination, and low contrast) leveraging several preprocessing algorithms. By utilizing graph-based segmentation, the authors' approach then leads to an innovation to isolate laser-fused materials from the unmolten powder bed. In the experiment of printing an overhang structure, this study provides detection of the formation of an overheating defect and proposes new features that correlate to part geometry and process parameters. These results can be further used to bridge the gap between spatially resolved process monitoring and ultimate model-based control for robust and high-throughput additive manufacturing.