A biological vision inspired framework for machine perception of abutting grating illusory contours

The article presents the illusory contour perception network (ICPNet), a novel deep learning architecture designed to enhance machine perception of abutting grating illusory contours by mimicking visual cortex circuits. Key features include a multi-scale feature projection (MFP) module for representation extraction, a feature interaction attention module (FIAM) for improved feature interaction, and an edge fusion module (EFM) that enforces shape constraints. Evaluated on the AG-MNIST and AG-Fashion-MNIST datasets, ICPNet demonstrates superior sensitivity to illusory contours, achieving significant improvements in top-1 accuracy compared to existing models, which may advance the alignment of DNNs with human perceptual capabilities.