摘要
用于计算图像分形维的差分盒计数法(DBC)和鲁棒差分盒计数法(RDBC)都对脉冲噪声和斑点噪声较敏感,为此本文提出一种抗噪差分盒计数法(NRDBC),利用剪切局部标准差(TLSD)来计算图像分形维,由于TLSD可有效滤除脉冲噪声和斑点噪声、且对高斯噪声敏感性小,因此NRDBC能对有噪图像进行可靠的分形维估计。利用多分辨率的DBC、RDBC和NRDBC对混有高斯噪声、脉冲噪声和斑点噪声的7种Brodatz纹理以及3种海底的侧扫声纳图像进行了分类实验,结果表明,本文提出的NRDBC可获得更高的识别率和更好的抗噪性。
The differential box-counting method （DBC） for calculating fractal dimension of an image has a disadvantage of being sensitive to noise. In addition, although the robust differential box-counting method （RDBC） can suppress Guassian noises effectively, it is sensitive to pulse and speckle noises. For this reason, a noise-resistant differential box-counting method （NRDBC） is here proposed for calculating fractal dimension of a noisy image, which uses a trimmed local standard deviation （TLSD） to compute the number of boxes needed to cover intensity surface of the image. TLSD has the characteristics of being capable of filtering pulse and speckle noises effectively and of being less sensitive to Guassian noises, and so the NRDBC can be used to calculate fractal dimension of noisy images reliably. Classification experiments on Brodatz texture images and side-scan sonar seafloor images are performed by using DBC, RDBC, and NRDBC respectively, all of which show that the NRDBC achieves higher classification rate and better noise resistivity.
出处
《应用声学》
CSCD
北大核心
2008年第2期95-101,共7页
Journal of Applied Acoustics
基金
中科院声学所所长择优基金项目(S2004-03)
关键词
分形维
差分盒计数法
鲁棒差分盒计数法
抗噪差分盒计数法
剪切局部标准差
Fractal dimension, Differential box-counting, Rrobust differential boxcounting, Noise-resistant Differential box-counting, Trimmed local standard deviation
