Texture Image Classification Using LLGMN and HLAC Features

In the manufacturing sector, there is a continuous demand for the high-quality, low-volume production of diverse products from the perspective of diversifying customer needs. The classification of products with complex textures is challenging. In this study, we classify artificially generated texture images using Log-Linearized Gaussian Mixture Networks (LLGMN), which are neural networks based on a probabilistic statistical model. This was achieved by inputting features obtained from the principal component analysis of Higher-order Local Auto-Correlation features (HLAC). Using the proposed method, we demonstrate that it is possible to quickly classify six classes of texture images with a high accuracy of 98 % and a training time of approximately 60 s on a standard PC. Based on these results, we anticipate potential applications in manufacturing sectors that require fast texture-image classification.