Towards Automated Perceptual Shot Matching in Motion Pictures

We present an automated shot-matching algorithm that incorporates perceptual effects and scene-specific context. Focusing on viewer-centric elements, such as attention to faces or objects and overall color mood, the system uses publicly available deep learning models to detect salient features. A committee-based approach matches these aspects independently, then combines them into a single RGB balance and black-level offset per shot. Although fully automated, the system supports manual adjustments to the strictness of face, object, mood, and temporal matching. This transparent, parameterized design avoids black-box behavior and provides intuitive, editable results for colorists. Visual examples demonstrate perceptual accuracy, and a study with five professional colorists evaluates match quality and potential time savings. Comparing manual grading of raw footage with algorithm-assisted pre-matched footage, results show a human-comparable match achieved 20% faster, with the algorithm reaching about 70% of colorist quality. These findings highlight the value of extended real-world evaluation.