基于子窗口字典学习的机织物纹理表征及应用

Texture Characterization and Application of Woven Fabric Based on Dictionary Learning with Sub-windows

为了提高机织物纹理表征算法的稳定性,提出了以子窗口字典学习表征机织物纹理的算法。将整幅图像划分为多个子窗口样本,并将子窗口样本展成列向量,所有的列向量联合组成灰度数据矩阵。选定离散余弦变换(discrete cosine transform,DCT)作为初始字典,对子窗口样本矩阵进行字典学习,最终得到了稳定的学习字典。选用均方根误差作为评价指标,对字典个数和子窗口大小进行优化。结果表明,应用学习得到的字典,不仅能近似重构机织物纹理样本图像,而且能在无监督的条件下自动识别织物的瑕疵。

The sub-window dictionary learning algorithm was proposed to improve the stability of the woven fabric texture characterization algorithm.The overall image was divided into small patches which were unfolded into column vectors, and the data matrix was formed by all the column vectors.The discrete cosine transform (DCT) was selected as the initial dictionary.Then we performed dictionary learning on the sub-window sample matrix and obtained the stable learned dictionary.The root mean square error was selected as evaluation index to optimize the size of dictionary and small patches.The experimental results demonstrate that the proposed algorithm can well represent fabric samples,and the dictionary can automatically indentify fabric defecTSwithout supervision.