Learning parallel and hierarchical mechanisms for edge detection

Edge detection is one of the fundamental components advanced computer vision tasks, and it essential to preserve computational resources while ensuring a certain level performance. In this paper, we propose lightweight edge network called Parallel Hierarchical Network (PHNet), which draws inspiration from parallel processing hierarchical mechanisms visual information in cortex neurons implemented via convolutional neural (CNN). Specifically, designed an encoding with based on transmission pathway “retina-LGN-V1” meticulously modeled receptive fields cells involved pathway. Empirical evaluation demonstrates that, despite minimal parameter count only 0.2 M, proposed model achieves remarkable ODS score 0.781 BSDS500 dataset 0.863 MBDD dataset. These results underscore efficacy attaining superior performance at low cost. Moreover, believe that study, combines biological vision, can provide new insights into research.