This paper focuses on digitization efforts for Black digital archives, particularly historical Black newspapers, which are structurally underrepresented in AI research and infrastructure. To address the challenges of accurate transcription due to inconsistent typography, visual degradation, and limited annotated layout data, this paper presents a layout-aware OCR pipeline tailored to Black newspaper archives and an unsupervised learning evaluation framework suitable for low-resource archival environments. Combining synthetic layout generation, model pretraining with augmented data, and state-of-the-art YOLO detector fusion, we evaluate a 400-page dataset of 10 Black newspaper titles using three unannotated evaluation metrics: semantic consistency score, region entropy, and text redundancy score. We demonstrate that layout-aware OCR improves structural diversity and reduces redundancy, with a slight trade-off in consistency.
