Multi-Scale LBP and GMM-Based Texture Recognition and Reproduction for Building Decoration Materials

Abstract

With the increasing demand for material texture recognition and reproduction in the architectural decoration industry, traditional methods relying on manual annotation are inefficient, and existing technologies are difficult to meet the requirements of complex texture processing. To accurately identify the texture and high-quality reproduction of building decoration materials, a texture recognition method based on multi-scale Local Binary Pattern algorithm combined with Gaussian mixture model clustering is proposed, and a complete model including image preprocessing, feature extraction, and texture reproduction modules is constructed. The experimental results on the Dresden Texture Dataset dataset show that the proposed method performs well in texture feature extraction, with a maximum accuracy of 96.84% for tile texture recognition and an average intersection to union ratio (MIoU) of 0.97. The practical application results show that the proposed method has a fast texture recognition speed and less memory consumption. The average time and size of the generated CSV files are 2.93 s and 8.75 KB, respectively, while the average time and size of the CSV files generated by the 3D Gaussian speckle model are 4.42 s and 21.41 KB, respectively. Meanwhile, the comprehensive score of texture reproduction performance is 8.93 points, followed by 8.27 points for 3D Gaussian speckle. From this, the method proposed by the research can quickly identify and reproduce the complex textures of building decoration materials with high fidelity. This study provides a new technological solution for digital texture recognition and production control of building decoration materials, promoting the intelligent and high-quality development of the building decoration industry.