We propose a combination of image processing methods to detect ceramic tiles defects automatically. The primary goal is to identify faults in ceramic tiles, with or without texture. The process consists of four steps: preprocessing, feature extraction, optimization, and classification. In the second step, gray-level co-occurrence matrix, segmentation-based fractal texture analysis, discrete wavelet transform, local binary pattern, and a novel method composed of segmentation-based fractal texture analysis and discrete wavelet transform are applied. The genetic algorithm was used to optimize the parameters. In the classification step, k-nearest neighbor, support vector machine, multilayer perceptron, probabilistic neural network, and radial basis function network were assessed. Two datasets were used to validate the proposed process, totaling 782 ceramic tiles. In comparison with the other feature extraction methods commonly used, we demonstrate that the use of SFTA with DWT had a remarkable increase in the overall accuracy, without compromising computational time. The proposed method can be executed in real time on actual production lines and reaches a defect detection accuracy of 99.01% for smooth tiles and 97.89% for textured ones.

我们提出一种图像处理方法的组合来自动检测瓷砖缺陷。主要目标是确定有无纹理的瓷砖缺陷。该过程包括四个步骤：预处理，特征提取，优化和分类。第二步，应用灰度共现矩阵，基于分段的分形纹理分析，离散小波变换，局部二进制模式以及由分段的分形纹理分析和离散小波变换组成的新方法。遗传算法被用来优化参数。在分类步骤中，k近邻，支持向量机，多层感知器，概率神经网络和径向基函数网络进行了评估。使用两个数据集来验证所提出的过程，总共782块瓷砖。与通常使用的其他特征提取方法相比，我们证明了将SFTA与DWT结合使用可显着提高整体准确性，而不会影响计算时间。所提出的方法可以在实际的生产线上实时执行，缺陷检测精度对于光滑瓷砖而言达到了99.01％，对于纹理瓷砖而言达到了97.89％。