基于奇异值分解的多尺度变换域图像细化算法  

A Thinning Algorithm of Multi-scale Transform Domain Image Based on Singular Value Decomposition

作　　者：沈丹萍[1] SHEN Dan-ping(Department of Computer Science and Technology/Suzhou College of Information Technology,Suzhou 215200,China)

机构地区：[1]苏州信息职业技术学院计算机科学与技术系,江苏苏州215200

出　　处：《山东农业大学学报（自然科学版）》2020年第2期262-265,共4页Journal of Shandong Agricultural University：Natural Science Edition

基　　金：“十二五”江苏省高等学校重点专业信息通信技术(198)。

摘　　要：由于多尺度变换域图像细化可降低图像存储难度与识别难度,图像细化核心即为骨架连通性,围绕保持图像骨架高度连通性这一宗旨,提出基于奇异值分解的多尺度变换域图像细化算法。首先,通过采用基于小波多尺度奇异值分解的图像特征提取算法,全面提取多尺度变换域图像中全部目标特征。然后,通过采用一种有效的图像细化算法判断图像细化时需留下的特征点,并去除图像中冗余像素点与冗余枝线,完成多尺度变换域图像细化。最后,根据本文所提算法将多尺度变换域图像以指纹、不规则图像等为例,对两幅图像细化,进行图像的细化测试。研究结果表明,所提算法细化后的图像纹理清楚,骨架分明,且图像纹络连通性较好;与同类型算法相比,该算法对多尺度变换域图像中冗余像素点和枝线的去除效果最佳。Because image thinning in multi-scale transform domain can reduce the difficulty of image storage and recognition, the core of image thinning is skeleton connectivity. To maintain the high connectivity of image skeleton, a multi-scale transform domain image thinning algorithm based on singular value decomposition is proposed. Firstly, the image feature extraction algorithm based on wavelet multi-scale singular value decomposition is used to extract all target features in multi-scale transform domain. Then, an effective image thinning algorithm is used to judge the feature points left behind in image thinning, and the redundant pixels and branches are removed to complete image thinning in multi-scale transform domain. Finally, according to the algorithm proposed in this paper, the multi-scale transform domain image is taken as an example of fingerprint, irregular image and so on. The two images are thinned and the thinning test is carried out. The results show that the refined image has clear texture, clear skeleton and good connectivity of image texture. Compared with the same type of algorithm, the algorithm has the best effect on removing redundant pixels and branches in multi-scale transform domain images.

关 键 词：奇异值 图像 算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]