A Biologically Inspired Contour Detection Model Based on Multiple Visual Channels and Multi-Hierarchical Visual Information

Abstract

The information processing mechanism of a visual system captures the real contours of an object through multi-hierarchical processing and integration of the basic features (color, grayscale, etc.) of the image. In more complex situations, a reasonable and effective framework is often required to solve the high-precision problem of contour detection. In this paper, we propose a contour detection model inspired by the biological characteristics of multiple visual channels and multi-hierarchical visual information. In our model, we mainly include information processing channels for color and luminance. The color channel response is obtained by color-sensitive neurons of the primary visual cortex (V1) processing the single-opponent response of the retinal/lateral geniculate nucleus (LGN) layer, and the luminance channel response is selected by the linear X cell response and the nonlinear Y cell response, based on the luminance contrast. The texture suppression of the color and luminance channel responses is enhanced by introducing a sparse coding mechanism. Finally, the fusion weights of the color and luminance channel response energies are calculated using the optimal direction to obtain better detection performance. The experimental results show that our model performs well in complex situations in comparison to current advanced biologically inspired model and some machine learning-based models, further improving the robustness of the non-learning contour detection model.