亚像素级数字图像弱边缘小目标快速检测算法  

Fast detection algorithm for small objects with weak edges in sub⁃pixel level digital images

作　　者：谢绍敏 李新荣 XIE Shaomin;LI Xinrong(School of Computer Engineering,Guilin University of Electronic Technology,Beihai 536000,China)

机构地区：[1]桂林电子科技大学计算机工程学院,广西北海536000

出　　处：《现代电子技术》2024年第13期23-26,共4页Modern Electronics Technique

摘　　要：小目标往往在图像中占据较少的像素区域,与背景色彩相近,导致对其检测面临目标分辨率低、目标与背景相似度高等问题,使得传统的像素级定位方法无法满足亚像素精度的检测需求。为此,文中研究亚像素级数字图像弱边缘小目标快速检测算法。基于三次B样条小波模极大值方法多层分解并计算图像弱边缘的模极大值,获取弱边缘细节图像,将其输入亚像素级的Franklin矩方法中,对其旋转处理,检测亚像素弱边缘目标点,采用最大类间方差法确定最佳灰度差阈值,提升图像弱边缘小目标快速检测能力。测试结果显示:图像信杂比增益均在0.017以下;背景抑制因子结果均在0.922以上;亚像素坐标计算结果和实际结果之间的误差均低于(0.11,0.13),清晰呈现了弱边缘小目标的分布情况。Small objects often occupy a small pixel area in the image.They are similar to the background color,which results in the low object resolution and high similarity between the objects and the background,which makes the traditional pixel-level positioning methods unable to meet the detection requirements of sub-pixel accuracy.Therefore,a fast detection algorithm for small objects with weak edges in sub-pixel digital images is studied.On the basis of the cubic B-spline wavelet modulus maximum method,the modulus maximum of the weak edge of the image is decomposed in multiple layers and calculated.The detailed image of the weak edge is obtained and input into the Franklin moment method at the sub-pixel level.The rotation process is performed to detect the object point of the weak edge of the sub-pixel.The maximum between-class variance method is used to determine the best gray difference threshold,so as to improve the ability of fast detection of small objects with weak edges in images.The test results show that the image signal-to-clutter ratio(SCR)gain is below 0.017,the results of inhibitory factors in the background are all above 0.922,and the errors between the calculated sub-pixel coordinates and the actual results are lower than(0.11,0.13),which clearly shows the distribution of small objects with weak edges.

关 键 词：亚像素级 数字图像 弱边缘 小目标 快速检测 模极大值 Franklin矩 灰度差阈值 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]