Abstract: In existing methods of ceramic tile online classification and detection, some kinds of texture are hard to quantify and it is also difficult to classify by color or gray scale. To solve the problem, first of all, the method based on Gaussian filter has been used to enhance texture and remove low-amplitude noise. And then, it is able to extract and process the features in frequency domain by the discrete fourier transform. Moreover, by removing the DC component and high-frequency components, texture has been highlighted and corresponding noise has been restrained. At last, it adopts minimum distance classification rules to compare every image with default template library. The result of experiments shows that it is very useful to distinguish various classification and robust in the same.

Key words: frequency spectrum, discrete Fourier transform, texture analysis, ceramic tile classification, minimum distance classification

摘要： 针对瓷砖在线分类检测中，一些瓷砖品种的纹理难以定量化描述，且种间色差、亮度差难以区分的问题，提出一种基于频率谱处理的瓷砖纹理分类方法。首先采用基于高斯滤波的方法增强纹理并消除低幅值噪声，从而加强了对低质量图像的适应性。然后通过离散傅里叶变换得到频率谱图像，去除直流分量和高频成分后既可以突出纹理信息又抑制了高频噪声的影响。最后与预设模板库比较，采用图像匹配法计算距离，利用最小距离分类法实现分类，满足对各种纹理瓷砖的适应性以及在线分类快速性要求。实验结果表明，该方法对不同的瓷砖样本区分度高，对同类样本鲁棒性好，分类准确率高，在瓷砖在线分类检测中具有较高的实用价值。

关键词: 频率谱, 离散傅里叶变换, 纹理分析, 瓷砖分类, 最小距离分类