基于神经元颜色拮抗与动态编码的轮廓检测方法

A Contour Detection Method Based on Color Opponent and Dynamic Coding of Neurons

基于神经元的颜色拮抗特性及神经元群体的动态编码机制,实现对图像的轮廓检测。模拟视皮层下神经元的颜色单拮抗特性,引入单拮抗感受野的动态调节机制,以充分响应颜色边界和亮度边界;利用单细胞的树突极性分布,构建初级视皮层的双拮抗神经元网络,实现对特定方位的视觉刺激响应,有效提取目标轮廓;在神经元的群体感受野内,考虑神经元的动态突触连接,融合单细胞的脉冲频率响应,实现对纹理信息的抑制作用。以BSDS500图库的图像为实验对象,结果显示该方法在提取主体轮廓的过程中能有效抑制纹理信息,其对100幅图像最佳检测结果的P值指标均值和标准差为0.58±0.04,相对CORF和CO等其他对比方法,可提高轮廓提取的准确率。所提出方法可有效实现图像的轮廓检测,为利用颜色信息以及神经元之间的动态编码、实现更高级皮层的图像理解或者视觉认知提供新的思路。

Based on characteristics of color opponent and dynamic coding mechanism of neuron population, to realize the contour detection of images. Modeling the single color opponent characteristics of neurons on sub- cortex, a kind of dynamic adjustment mechanism about receptive field with single color opponent was constructed, in order to respond sufficiently on both color and lightness boundaries. Using the dendrite polar distribution of single cells, a neuron network with double color opponent on primary cortex was built to detect visual stimulus of specific orientation, in order to effectively extract the contour. Finally, taking the dynamic synaptic link into consideration, spike frequency response of single cells was synchronized in receptive field of neuron population, in purpose of realizing the inhibition of texture information. BSDS500 database was used in the experiments, the result indicates that the proposed method can effectively inhibit the texture information in the process of extracting the contour, the mean value and standard deviation of measure P for 100 images was 0.58 ± 0.04, relative to other contrast method, improves the accuracy of contour extraction. Our method is effective for image contour detection, and provides a new idea for the image coding or visual cognition of the higher cortex by using the color information and the dynamic coding between the neurons.