Abstract: The segmentation of Miao costume patterns not only contributes to the protection and transmission of the unique Miao culture, but offers a wealth of materials and inspiration for artistic creation, brand promotion, and academic research. At the same time, through the segmentation of Miao costume patterns, the Miao history, culture and tradition can be better understood and spread, and the sense of national identity and cultural self-confidence can be enhanced. In addition, it contributes to digital preservation, providing new ideas and methods in the field of modern technology and the segmentation of national costumes. To address the poor robustness and noise sensitivity of the intuitive fuzzy clustering algorithm, a novel algorithm for the segmentation of Miao costume patterns was proposed, and it incorporated complementary spatial information and membership modification. Firstly, to enhance noise resistance, the algorithm innovatively employed a weighted squared Euclidean distance based on complementary spatial information instead of the traditional Euclidean distance calculation. In this way, the algorithm is able to more accurately measure the similarity between pixels, and effectively resist the interference of noise, thus achieving more accurate segmentation in complex and noisy images data. Secondly, to further improve the algorithm's efficiency, a membership connection mechanism was introduced. The core idea of this mechanism is to optimize the computation and update process of the membership functions, effectively reducing the number of iterations and ensuring robust performance even in resource-constrained environments. Finally, in order to achieve more accurate images segmentation, the membership function is corrected by utilizing the local pixel features and spatial relationships within the images. By assigning different weights to the pixels within the neighbourhood, the algorithm better captures the image's details and structural information, resulting in higher segmentation accuracy. This correction strategy considers both the local features of pixels and the spatial relationships between them, ensuring that the segmentation results more closely match the actual content of the images. When a mixed noise density was 10%, the proposed algorithm achieved a segmentation accuracy of 99.72% on the synthetic image dataset and obtained partition coefficient and partition entropy values of 97.23% and 4.61% on the Miao costume patterns dataset. Experimental results show that the proposed algorithm performs well in terms of segmentation accuracy and detail retention ability, and is significantly better than other related algorithms. When faced with noisy and colorful Miao costume patterns datasets, the algorithm accurately identifies and segments different regions within the images while retaining more detailed information.

Key words: intuitionistic fuzzy clustering, Miao costume, segmentation, noise, complementary spatial information

摘要： 苗族服饰图案分割对推动苗族服饰文化的数字化保护和传承具有重要意义。针对直觉模糊聚类算法鲁棒性差、对噪声敏感的问题，提出一种基于互补空间信息和隶属度修正的直觉模糊聚类苗族服饰图案分割算法。首先，该算法使用互补空间信息的加权平方欧式距离代替传统欧氏距离，用于提高算法的抗噪性能；其次，采用隶属度连接机制，减少算法的迭代次数，从而提升算法的运行速率；最后，利用图像的局部像素特征和空间关系，对邻域内的像素点赋予不同的权重来修正隶属度函数，以实现更为准确的分割。当混合噪声的密度为10%时，所提算法在合成图像数据集上的分割精度达到99.72%，在苗族服饰图案数据集上的划分系数和划分熵为97.23%和4.61%。实验结果表明，与相关算法相比，所提算法的分割精度更高、细节保留能力更强。

关键词: 直觉模糊聚类, 苗族服饰, 分割, 噪声, 互补空间信息