【摘要】 针对图像轮廓提取存在不完整的问题，受生物视觉信息处理机制启发，提出一种图像轮廓提取的新方法。首先，模拟视网膜水平细胞的光感受特性，构建亮度增益模型，突出图像细节特征；其次，模拟M通路亮度信息处理过程，结合固视微动特性和神经元动态编码过程，构建亮度对比度特征提取模型抑制纹理信息；同时，模拟P通路中心分离感受野的颜色信息处理机制，提取彩色轮廓信息；最后，模拟初级视皮层V1区的特征信息融合机理，整合M通路和P通路平行通道信息，最终实现对图像轮廓特征的完整提取。针对BSDS500数据集和BIPED数据集，最优平均P指标分别为0.42和0.50,表明所提方法能够有效地抑制纹理噪声以提取轮廓特征。基于生物视觉信息处理机制提出的轮廓处理模型，可以较好地提取图像完整的轮廓并具有可解释性强，为面向图像处理或理解的视觉机制应用提供了一种新思路。更多还原

【Abstract】 Inspired by the biological visual information processing mechanism, a new method for image contour extraction is proposed to address the issue of incomplete contour extraction. Firstly, a brightness gain model is constructed to simulate the light sensing characteristics of the retinal horizontal cells, which enhances the image detail features. Secondly, a brightness-contrast feature extraction model is constructed to suppress texture information by simulating the processing of brightness information in the M pathway, combining the properties of fixational eye movements and dynamic neural coding. Additionally, the color contour information is extracted by simulating the color processing mechanism of the center-surround receptive fields in the P pathway. Finally, the parallel channel information from the M and P pathways is integrated by simulating the feature information fusion mechanism in the V1 primary visual cortex, to achieve complete contour feature extraction. The proposed method is evaluated on the BSDS500 and BIPED datasets, and the best average P indicators for a single image are 0.42 and 0.50,respectively, indicating that the proposed method is effective in suppressing texture noise and extracting contour features. Based on the biological visual information processing mechanism, the proposed contour processing model can extract image contours completely and has strong interpretability, providing a new approach for the application of visual mechanisms in image processing or understanding.更多还原