基于梯度二阶导数的Canny阈值自适应选取算法

图像边缘含有丰富的图像信息,对于很多视觉系统至关重要,比如目标检测与图像分割等。传统的Canny算子仅能用于灰度图像,无法有效利用彩色图像中的颜色信息;此外,该方法需要人为设定高低阈值,不能自适应提取图像边缘,进而造成背景边缘放大等问题。提出了一种基于Canny的自适应彩色图像边缘检测算法;该算法首先通过图像一阶直方图的统计特性,自适应地选取高斯滤波器的参数σ,有效去除了噪声;同时改善了σ参数设置不合理对边缘检测的影响。其次采用了基于图像梯度二阶导数的阈值选取的方法,即根据图像特性自适应选取合适的阈值。实验结果表明:算法能很好地改善传统Canny算子的缺陷,对于彩色图像的边缘信息提取能达到很好的效果。

The Canny threshold selection algorithm based on the second derivative of image gradient

Image edges contain abundant information, which is crucial for many systems, such as target detection and image segmentation. Traditional Canny can only be used for gray images, which cannot effectively use color information in multicolor images. In addition, the method needs to set the high and low threshold, which cannot extract edges adaptively, and can cause other problems such as amplification of background edges. This paper presents an adaptive edge detection algorithm based on Canny in multicolor images. Firstly, this algorithm adaptively selects parameters of the Gaussian filter by first-order statistical properties of the histogram of images, which effectively removes the noise and reduces the influence of unreasonable parameter setting on edge detection. Secondly, the threshold selection method based on the two derivative of image gradient is adopted to adaptively select the appropriate threshold according to characteristics of images. Experimental results show that the proposed algorithm can improve defects of traditional Canny operator, and can extract edge information from multicolor images effectively.