A professionally developed visual coding framework for affective video analysis

Introduction Affective responses to videos arise from multidimensional, context-dependent visual features that unfold over time. However, no standardized method exists to characterize these features, limiting research on visual-based immersive interventions and broader computer vision science. We developed an expert-informed visual coding framework to provide a systematic approach to visual coding. Methods A visual coding framework was developed by film professionals to characterize cinematographic features that may be relevant to viewers’ emotional responses. The framework was initially evaluated using selected nature-based landscape clips from a publicly available database of 360° immersive videos. Results The visual coding framework contains 17 items. We evaluated this framework using a set of 48 natural landscape video clips to represent a broad range of environmental characteristics, including forests, mountains, beaches, and deserts. Across a total of 816 unique video-item evaluations, the mean modal agreement was 0.720, indicating 72.0% of the independent film professionals converged on the exact same dominant response category for any given visual feature. Mean exact pairwise agreement across items was 0.584, and item-level Krippendorff’s alpha ranged from 0.043 to 0.463. The differences between raw agreement and alpha reflect prevalent video characteristics and limited video stimulus variability due to the exclusive use of selected natural-landscape videos. Conclusion This framework provides a systematic approach to characterizing video stimuli and establishes a foundation for identifying visual features of immersive nature videos that may influence psychological and behavioral outcomes.