Submarine Periscope Artifact Real-Time Correction (SPARC)

The Navy has promising efforts underway to automate 3D model building and identification of ships from submarine periscope video. Currently, this video contains artifacts due to reticle lines, lens smudges, interlacing, and color sampling. These artifacts must be must be removed by hand, a time-consuming preprocessing effort, before being fed to downstream video interpretation processes. Stellar Science proposes to demonstrate the feasibility of removing such artifacts in real-time by developing and integrating algorithms for deinterlacing, reticle line and smudge removal, and Bayer demosaicing. Our approach to video preprocessing is motivated by our understanding of downstream video interpretation algorithms. It is based on the physics of relative motion and image formation, in order to recover scene irradiance, rather than the infilling of pixels to make a visually pleasing image. By formulating all types of artifact correction as a single-pass, weighted resampling operation, we will achieve real-time throughput with high fidelity to the original scene. Having recently developed an initial implementation of a system for building 3D ship models from video, we will test and demonstrate that our automated preprocessing algorithms satisfy the image quality requirements of downstream processes.