摘要
文中提出一种基于外轮廓模糊处理的多尺度目标检测方法。由于目标背景区域通常与图像边界相连接,因此,文中通过计算与图像边界的距离提取显著目标,且利用超像素过分割提高处理效率。具体而言,首先对图像进行超像素分割;然后,依据超像素与图像边界距离生成最小树,并以此获得显著目标的初步检测结果;接下来,利用快速轮廓检测法提取显著目标的外轮廓信息;最后,利用模糊色差直方图及多尺度方法获得显著目标的准确检测结果。实验结果表明,与现有方法相比,文中所提算法在效率和精度上具有一定优势。
A multi-scale target detection method based on the fuzzy treatment of image skeleton is presen- ted. Since the background region of each salient object is connected to the image boundary, each salient object can be detected by computing the distance to the image boundary, and the super-pixel segmentation is utilized to improve the computation speed. Firstly, the simple linear iterative cluster method is used to segment each image into super-pixels. Secondly, the distance between each super pixel and the image boundary is calculated to generate the minimum tree to obtain the initial result of each salient object. Then, the fast skeleton detection method is utilized to obtain the skeleton information. Finally, the fuzzy chromatism histogram is adopted to obtain accurate detection result of each salient object. Experimental results on benchmark datasets demonstrate that, compared with the existed methods, the method enhances the superiority of effectiveness and accuracy.
作者
程艳云
朱松豪
石路路
CHENG Yanyun;ZHU Songhao;SHI Lulu(School of Automation, Nanjing University of Posts and Telecommunications, Nanjing 210023, China;Department of Research and Development, Nanjing Huasu Technology Corporation Limited, Nanjing 210009, China)
出处
《南京邮电大学学报（自然科学版）》
北大核心
2018年第2期78-86,共9页
Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition
基金
南京邮电大学国自基金孵化项目(NY217066)资助项目
关键词
显著目标检测
最小生成树
外轮廓提取
模糊色差直方图
多尺度检测
salient object detection
minimum tree
skeleton extraction
fuzzy chromatic aberration histo- gram
muhiscale detection
