Contour detection model inspired by V1 surround modulation

Contour detection plays an important role in visual perception tasks and is key in subsequent target detection and recognition. The existing contour detection algorithms have achieved relatively good results. Yet knowledge gaps remain concerning the insufficient texture suppression and missing subject contour. Physiological studies have shown that the human visual system can effectively capture edge features of input images via neurons in the primary visual cortex (V1). Therefore, a contour detection model based on surround modulation in V1 is proposed. Firstly, to achieve an effective balance between texture suppression and contour extraction, an adaptive surround modulation model is proposed. Then considering the diversity of contour features in the image, combined with the multi-scale structure characteristics of the receptive field itself, the multi-scale surround modulation mechanism model is introduced to reduce the texture redundancy information. Choosing the BSDS500 natural scene dataset as the experimental object, the F-Score is selected as the evaluation index. The average optimal F-Score value of the proposed method is 0.703, which is higher than other biological vision-based methods. Further tests on NYUD data show that the model has good generalization. The proposed adaptive surround modulation model can effectively solve the imbalance between texture suppression and contour extraction and provide a new insight for subsequent image processing tasks.