热红外影像特殊的成像机理和特点, 使影像中存在大量的噪声导致边缘信息模糊难提取。基于传统 Sobel算子提出了一种新的边缘检测方法。首先利用数学形态学中的不同尺度、不同结构元素对具有随机噪声的热红外影像进行形态学去噪, 再用 Otsu算法对去滤波后的影像进行全局阈值分割, 最后利用 Sobel算子对其进行边缘提取。基于 MATLAB仿真实验, 结果表明: 与传统的 Sobel算子相比, 该方法不仅有较强的抗噪性, 而且检测出的边缘外部轮廓与内部细节特征表达较好, 边缘具有连续一致性, 是一种简单、快速的边缘检测新方法。

There is so much noise caused by special imaging mechanism in infrared thermal images, which leads to difficulty in extracting edge. A new edge detection method was proposed based on traditional Sobel operator. Different scales and structure elements of mathematical morphology were applied to filter infrared thermal images with random noise and global threshold algorithms for image segmentation (Otsu algorithm) was used to classify images. Finally, edge information was extracted by Sobel operator. Based on MATLAB program, the result shows that the new method not only has strong noise immunity, but also has a better detection effect on outer and internal features compared with traditional Sobel operator. It has a better continuous consistency and it is a simple and fast edge extraction method.