The pattern of spatial contrast discontinuities in natural images has been analysed in the present work, and based on it, a new adaptive model of the bio-inspired Difference of Gaussian (DOG)-based edge detector has been designed. The distinguishing feature of the proposed filter is that the magnitude of surround suppression in receptive field of the DOG is adaptively adjusted depending on the nature of discontinuity of the edge profile. The model is based on the biological evidences indicating the possibility that human brain may be endowed with the ability to perform Fourier decomposition of visual images into its various components of spatial frequencies. It may be shown that information obtained from such a Fourier decomposition may help to measure the strength of contrast (sharpness of discontinuity) in the intensity profile across any possible edge in the natural image. In the present model, it is assumed that the magnitude of surround suppression in an excitatory–inhibitory receptive field is dependent on the sharpness of discontinuity. The suppression is strong when the edge contrast is poor, while it becomes weaker as the edge contrast is high. At a biphasic edge, the surround suppression is vanishingly small. Natural images collected from benchmark databases are used to evaluate the efficiency and robustness of the proposed model for the detection of edges. The result shows that the edge maps generated through the proposed model are at par, if not more effective as compared to the classical edge detectors like Canny. The performance of the proposed model is also compared with a number of recently proposed alternative adaptive models for edge detection.

在当前工作中分析了自然图像中空间对比度不连续性的模式，并在此模式的基础上，设计了基于生物启发性高斯差分（DOG）的边缘检测器的新自适应模型。所提出的滤波器的区别特征在于，根据边缘轮廓的不连续性来自适应地调节DOG的接收场中的环绕抑制的幅度。该模型基于生物学证据，该生物学证据表明可能赋予人类大脑对视觉图像进行傅立叶分解为空间频率各个组成部分的能力。可以表明，从这种傅立叶分解获得的信息可以帮助测量自然图像中任何可能边缘的强度分布中的对比强度（不连续的清晰度）。在本模型中，假设在兴奋性抑制性感受野中周围抑制的幅度取决于间断的锐度。当边缘对比度差时，抑制作用很强，而当边缘对比度较高时，抑制作用就变弱。在双相边缘，环绕声抑制非常小。从基准数据库收集的自然图像用于评估所提出的模型用于边缘检测的效率和鲁棒性。结果表明，通过所提出的模型生成的边缘图是同等的，与经典的边缘检测器（如Canny）相比，效果甚至更好。还将所提出的模型的性能与许多最近提出的用于边缘检测的替代自适应模型进行比较。