Abstract: Aimed at the problems of the amount of computational load and low accuracy of Gabor wavelet for fabric defect detection, a new optimal objective function which is variance per unit mean is proposed. Fabric defect image with a multi-channel Gabor wavelet is filtered. The mean and variance of each filtered defect image are calculated. The rate between the variance and mean of each filtered image is calculated respectively. The maximal value of channel is selected as the optimal channel. The characteristics of fabric defect are almost either warp or weft defect. On the basis of the optimal objective function, an automated inspection system for fabric defect based on the optimal 1×2 2D Gabor wavelet is presented. Experimental results show that it lowers the computational load, and the accuracy is as high as 93%.

Key words: Gabor, fabric defect, optimal function

摘要： 针对基于Gabor小波的布匹瑕疵检测运算量大，准确率较低等问题，提出了一种新的优化选择算法，即均值归一化方差算法。布匹瑕疵图像通过多通道Gabor小波；分别计算各通道滤波图像的均值和方差；计算方差与均值的比值，选择比值最大的通道作为优化选择通道。考虑到布匹瑕疵主要表现为经向与纬向瑕疵，因此基于该算法设计了1×2二维Gabor小波布匹瑕疵自动检测系统，降低了运算量，提高了准确率，达到了93%。

关键词: Gabor, 布匹瑕疵, 优化选择