Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

Distinct Computational Mechanisms Underlie Holistic Processing of Faces and Non-Face Line Patterns

Elaheh Akbari-fathkouhi¹, Katharina Dobs²; ¹Justus Liebig University Gießen, ²Justus Liebig Universität Gießen

Presenter: Elaheh Akbari-fathkouhi

Holistic processing is the tendency to perceive objects as unified wholes. A hallmark is the composite effect—combining the top half of one object with the bottom half of another creates a novel percept that disappears when misaligned. Although traditionally considered face-specific or expertise-based, recent findings show that even unfamiliar line patterns can be processed holistically, raising the question: Do these processes rely on similar or distinct mechanisms? To find out, we used three convolutional neural networks (CNNs)—one trained on object categorization, one trained on face identification and one untrained—and tested them with faces and line patterns, mirroring human studies. The composite effect for faces emerged only in the face-trained CNN and was disrupted by inversion, suggesting a face-specific mechanism. In contrast, line patterns elicited a composite effect across trained CNNs regardless of inversion, pointing to a domain-general process. Notably, holistic processing for faces peaked at later processing stages than for line patterns. Our results suggest that distinct mechanisms underlie holistic perception for faces and line patterns in CNNs and, we conjecture, also in the human brain.

Topic Area: Object Recognition & Visual Attention

Extended Abstract: Full Text PDF