结合亮度加权的自适应分区间边缘优化提取方法  

Adaptive inter-partition edge extraction method combined with brightness weight

作　　者：徐辛超[1] 刘明岳 徐彦田[2] 赵红喜 田飞飞 XU Xinchao;LIU Mingyue;XU Yantian;ZHAO Hongxi;TIAN Feifei(School of Geomatics,Liaoning Technical University,Fuxin,Liaoning 123000,China;Chinese Academy of Surveying and Mapping,Beijing 100036,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [2]中国测绘科学研究院,北京100036

出　　处：《测绘科学》2022年第6期65-77,共13页Science of Surveying and Mapping

基　　金：辽宁省教育厅基础研究项目(LJ2019JL021);辽宁省自然科学基金项目(20180550849);地球观测与时空信息科学重点实验室项目(201901);国家自然科学基金项目(42071447)

摘　　要：针对传统边缘提取方法确定阈值时以图像整体信息为参考,导致梯度较弱的边缘无法自动提取的问题,该文提出了一种结合亮度分布的自适应分区间边缘提取方法。首先,采用优化梯度模板计算各像素梯度;其次,根据像素梯度方向对各像素3×3邻域亮度进行重采样确定像素的最终亮度值;然后,拟合像素梯度方向规律按照1倍标准差进行区间划分,并根据各区间像素亮度和梯度确定区间梯度阈值;最后结合区间梯度阈值进行边缘点筛选及连接,得到图像最终边缘。实验结果表明,与经典的自适应坎尼(Canny)边缘提取结果相比,提出的方法提取的边缘整体连续性更好,对图像中梯度较弱的边缘优化提取效果更加显著,均方误差平均降低了1.93%,峰值信噪比平均升高了0.44%。In order to solve the problem that the traditional edge extraction method taked the whole image information as a reference when determining the threshold, and the edge with weak gradient couldn’t be automatically extracted, this paper proposed an adaptive edge extraction method based on brightness distribution. Firstly, the optimized gradient template was used to calculate the gradient of each pixel. Secondly, resampling the 3×3 neighborhood brightness of each pixel according to the pixel gradient direction to determine the final brightness value of the pixel;Then, the fitting pixel gradient direction rule was divided into intervals according to one standard deviation, and the interval gradient threshold was determined according to the brightness and gradient of pixels in each interval. Finally, the edge points were screened and connected by interval gradient threshold, and the final edge of the image was obtained. The experimental results showed that, compared with the classical adaptive Canny edge extraction results, the proposed method had a better overall continuity of edge extraction, and had a more remarkable effect on edge extraction with weak gradient in the image. Mean-square error decreased by 1.93% on average and peak signal-to-noise ratio increased by 0.44% on average.

关 键 词：边缘提取 高斯分布 亮度 梯度方向 分区间 自适应 

分 类 号：P237[天文地球—摄影测量与遥感]