A bio-inspired edge and segment detection method by modeling multiple visual regions

Edges are critical in facilitating visual tasks, such as image segmentation and object recognition. The human visual system perceives edges through numerous cells that exhibit selective responses to oriented edges in the primary (V1) and secondary (V2) visual cortices. In light of this, many bio-inspired edge detection methods have been designed. However, how to extract object edges from texture backgrounds and connect these edges to form one-pixel-wide continuous edge segments is still challenging for them. In this work, a novel bio-inspired edge and segment detection method is proposed. The orientation-sensitive cells in V1 are modeled by Gaussian gradient functions to detect oriented edges. Their responses are modulated by a surround modulation mechanism and integrated with texture boundary cues to enhance object edges and suppress texture edges. Furthermore, end-stopped cells in V2 are innovatively modeled to connect edges detected in V1 and generate one-pixel-wide continuous edge segments. Finally, these edge segments are fed back to V1, resulting in enhanced edge detection results. Evaluations on three benchmark datasets demonstrate that the proposed method achieves the state of the art among bio-inspired methods while producing cleaner and more precise edge segments. The source code is available at https://github.com/DaipengYang7/BESD.