一种加权的多重分形特征提取算法  

Weighted multi-fractal algorithm for feature extraction

作　　者：谢雅婷[1] 梁光明[2] 石跃祥[1] 柳佳雯[1] 丁建文 

机构地区：[1]湘潭大学信息工程学院,湖南湘潭411105 [2]国防科技大学电子科学与工程学院,长沙410075 [3]长沙爱威科技,长沙410013

出　　处：《计算机工程与应用》2013年第4期206-208,共3页Computer Engineering and Applications

基　　金：湖南省自然科学基金(No.07JJ6115);智能制造湖南省高校重点实验室(No.2009IM06)

摘　　要：为解决多重分形维数不能够很好地反映图像强度信息和对图像尺度有强依赖的问题,在研究q阶广义维数D(q)基础上,提出两种改进方法。通过分析影响生长概率的因子,提出一种结合强度信息的加权子数计算方法,提出一种基于网格强度与均值的二维分形维数计算方法。实验表明改进的多重分形算法提了特征区分度,计算特征更加鲁棒和有效,将改进方法用于血细胞识别系统,改善了白细胞分类准确性。In order to solve the drawbacks of the multi-fractal dimension can not be a good reflection of the image intensity in- formation and highly dependes on the image scale, this paper presents two improvement based on q-order moments of general di- mension theory, marked D （q）. This paper proposes a new weighted calculation of boxes number method conbined intensity in- formation by analyzing the factors affecting the probability of growth, then proposes a two-dimensional method of calculating fractal dimension based on the gride intensity and mean. Experiment shows that the new method improves features differenta- tion, computes features more robust and more effective, and improves the classification accuracy by putting the new method in the identification systerm of blood cells.

关 键 词：特征提取 多重分形 网格 生长概率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]