一种基于人类视觉特性的边缘检测改进算法

针对传统Canny算子高低阈值的确定问题进行了研究。提出了一种基于人类视觉特性的Canny边缘检测算法。用频谱包络滤波器对图像进行平滑预处理,根据仿生学原理,分析了韦伯比曲线,将图像根据背景亮度划分为不同的区域,采用不同的公式计算高低阈值的方法。仿真实验表明,该方法在平滑图像时保留了重要的边缘信息,能够根据边缘所处背景亮度自适应地选择阈值,检测出与人的主观视觉更为一致的图像边缘。提高了边缘检测的自动化程度,更强的实用性。

Improved Canny edge detection based on human visual model

The traditional threshold of Canny operator is conducted a study to determine. Spectrum envelope is proposed for image smoothing filter. Through analyzing Weber’s law according to bionics principle, based on the background brightness the image is divided into different regions. The edge thresholds for the different areas are calculated according to the different equations. Experimental results show that the method can well protect the edge information of original image when the background noise is smoothing, and the edge thresholds can be adaptively selected according to the located backgrounds. Thus the image edge that is more consistent with human subjective vision can be detected. This method improves the automation of edge detection and is more practical.