摘要
研究了两类小波域图像纹理特征提取方法在非织造材料外观质量描述中的应用。分别从高频子带小波系数中计算1-范数和2-范数能量基特征,以及根据小波系数服从广义高斯分布,采用极大似然估计法计算广义高斯分布的尺度参数和形状参数作为非织造材料纹理特征。以1-紧邻分类器正确识别率为评价指标,衡量了两类小波纹理特征在非织造材料外观质量识别中的纹理表达能力和可分性。实验数据表明,提出的两类纹小波纹理特征在非织造材料外观质量识别中具有较强的刻画能力和较好的质量。
The research of the application of wavelet texture analysis on the visual quality description ofnonwovens is presented. The extraction method for two types of wavelet textural features is proposed, i.e., one is to compute the 1-norm and 2-norm values of wavelet coefficients as energy-based textural features, the other is to estimate the scale and shape parameters of generalized Gaussian distribution that fits the wavelet coefficients histogram well, and the two parameters are used jointly as wavelet texture features. To assess the de- scription capacity and separability of the wavelet texture features in the identification of visual quality of nonwovens, the recognition ac- curacy of 1-nearest neighbor classifier is used as evaluation criterion. Experimental data indicates that the two types of wavelet texture features have powerful description ability and excellent quality in the recognition of visual quality of nonwovens.
出处
《计算机工程与设计》
CSCD
北大核心
2011年第8期2836-2840,2856,共6页
Computer Engineering and Design
基金
现代丝绸国家工程实验室开放课题基金项目(2010007)
关键词
小波纹理特征
纹理特征评价
广义高斯分布
形状参数
K-紧邻分类器
wavelet texture feature
texture feature evaluation
generalized Gaussian distribution
shape parameter
k-nearest neighbors classifier
