Abstract: Texture feature extraction is the key to texture classification, and there are various factors such as rotation, illumination, and scale variations in texture images. To enhance the robustness of the texture feature extraction algorithm for rotation, illumination, and scale variations, the FWLMP ( Fractal Weighted Local Morphological Pattern) is proposed. First, a scale-invariant descriptor is constructed by using the relative invariance of fractal dimensionality to scale variation. Then, it is sampled and analyzed using the expansion, erosion, and opening-closing operations in mathematical morphology, and its weights are calculated by using the fractal dimension image. This algorithm is scale-invariant and robust to rotation and illumination changes. To achieve the classification of Qing Dynasty costume images, the Qing Dynasty Buzi image dataset is constructed. The FWLMP and similar algorithms are tested on four public texture datasets and a private dataset constructed by ourselves. The experimental results show that the FWLMP algorithm performs well in texture image classification and in Buzi image classification for Qing Dynasty civil and military officials. 

Key words: fractal dimension, local morphological pattern, qing dynasty costume, dataset, texture classification