基于K-SVD学习字典的机织物纹理表征及应用

为更好地适应机织物纹理以及减少程序的运行时间，选取平纹、斜纹、缎纹3种组织结构采用Ｋ?奇异值分解（K-SVD）的方法训练得到一个自适应字典。以峰值信噪比、结构相似性为指标，探讨不同稀疏基数对机织物纹理图像重构的影响，针对不同的应用，选取了合适的稀疏基数T。利用该字典重构机织物纹理图像，在此基础上检测织物瑕疵。实验结果表明：T=6 时，算法不仅能有效重构机织物纹理图像（PSNR和SSIM），而且重构效果要优于初始离散余弦转换（DCT）字典；T=4 时，K-SVD 字典能更好地适应瑕疵样本，且鉴别瑕疵的能力更强。

Woven fabric texture representation and application based on K-SVD dictionary

In order to wel adapt the woven fabric texture and reduce the algorithm running time, three basic weave patterns (plain, twill and satin) were chosen as trained samples to learn an adaptive dictionary by K-means singular value decomposition (K-SVD) dictionary learning approach. In order to select appropriate sparsity cardinality T for different applications, peak sognal to noise ratio (PSNR) and structural similarity index measurement (SSIM) wer chosen as evaluating preformance indexes. For regular fabric texture image reconstruction, T=6, the experimental results demonstrate that the proposed method not only can approximate fabric samples well, but also can improve the quality of reconsturcted image (in terms of PSNR and SSIM), in comparison with discrete cosine transformation dectionary. In addition, for fabric flaw detection, T=4, the K-SVD can well adapt samples with defects, and has stronger capability of identifying defects, compared with discrete cosine transformation dictionary.