Divergent Roles of Visual Structure and Conceptual Meaning in Scene Detection and Categorization

Human observers can recognize the meaning of a complex visual scene in a fraction of a second, but not all scenes are equally easy to recognize at a glance. What governs this variability? We tested the hypothesis that scene understanding is modulated by two distinct forms of information: visual information, defined as the structural complexity of the image, and semantic information, defined as the richness of the scene’s conceptual content. We quantified visual information using image compressibility and quantified semantic information from the complexity of human-written scene descriptions. Across four behavioral experiments, participants performed either a rapid detection task (distinguishing intact scenes from phase-scrambled masks) or a basic-level categorization task. High visual information impaired both detection and categorization, consistent with a perceptual bottleneck. In contrast, high semantic information facilitated detection but not categorization, suggesting that conceptual richness facilitates early perceptual processes without necessarily improving recognition. These findings reveal a dissociation between visual and semantic scene attributes and suggest that top-down expectations can selectively support early perceptual processing.