A Contour Detection Method Based on the Projective Coding Model of the Visual Cortex Information Flow

Abstract

Based on the information flow partition projection characteristics of P-type and M-type ganglion cells (referred to as P cells and M cells) and the neural sparse coding mechanism, this paper proposed a new method for image contour detection. First, we considered the difference between M cells and P cells in detail sensitivity and the information transmission of different visual signals. The parallel visual pathway was constructed to simulate the pre-level characteristics of the V1 layer to obtain the primary contour response. Then, we introduced the orientation sensitivity and stimulus response difference of the visual receptive field to construct the visual information difference enhancement model. In consideration of the visual attention mechanism, we proposed an adaptive size sparse coding network model that simulates the pre-level characteristics of the V4 layer to intelligently focus the target contour features. At the same time, de-redundancy was performed to obtain the fine feature image. Finally, the hierarchical information feedback fusion was built, and the fine feature image was used to correct the primary contour response to obtain complete contour detection results. Taking the BSDS500 dataset as experimental objects, the results showed that the proposed method exhibits an effective trade-off between contour extraction and texture suppression.