织物表面折皱的小波分析与自组织神经网络等级评定

Wavelet Analysis of Fabric Surface Wrinkle and Self-organized Neural Network Grade Assessment

为了提取较为精细的图像信息,引入了多尺度2维小波分析织物的表面折皱。织物图像首先经过高斯滤波,再利用小波变换分解并从中提取高频信息,然后结合4种表征织物折皱的特征参数,计算不同折皱等级模板的特征值,通过分析特征值与折皱等级的相关系数,表明这4种特征参数可以作为模式识别的输入量;最后采用Kohonen自组织神经网络客观评定织物的折皱等级,自组织神经网络将不同等级的织物折皱模板进行分类,并以此为依据,确定26种不同织物类型的折皱等级。为了定量描述评定结果,通过计算客观评定与主观评定结果的相关系数,验证该方法的可行性。

In this paper, Multi-Scale two-dimensional wavelet transform is imported to analysze fabric surface wrinkle in order to acquire the finer image information. Firstly, fabric image can be filtered through Gaussian filter, and decomposed by wavelet transform; meanwhile, high frequency information is extracted. Secondly, four kinds of wrinkle feature parameter are applied to calculate the fabric wrinkle degree with different wrinkle replica, which are horizontal variance, vertical variance, horizontal offset and vertical offset separately. Through analyzing the correlation coefficient between feature parameter and wrinkle grade, which indicates the four kinds of wrinkle feature parameter can be taken as the input value for pattern recognition. Finally, Kohonen self-organized neural network is also used to evaluate fabric wrinkle grade objectively. The wrinkle feature parameters are input to the Kohonen self-organized neural network, through training and studying process, the output value can be obtained, different wrinkle grade of fabric replica will be classified by applying self-organized neural network, and wrinkle grade of 26 different type fabrics can be evaluated according to this result. For describing the assessment result with quantify, the correlation coefficient is calculated between objective assessment and subjective assessment in order to validate the feasibility of this method.