印花织物自动分色系统

摘要

为了区分印花织物上彩色图案的色彩区，满足行业对织物色彩的需求，提出一种 SOM 神经网络与改进型 K-均值算法相结合的印花织物自动分色系统。在三维 RGB 颜色空间内，利用遗传算法搜寻出代表原织物颜色分布的子图像，并采用两层聚类方法完成分色：第一层使用 SOM神经网络对大于期望聚类数的数据样本进行初始聚类；第二层对 SOM 神经网络初始聚类的神经元利用改进型 K-均值算法进一步聚类，形成最终的聚类结果，从而完成分色。实验结果显示，此系统能够准确地区分出印花织物图案的色彩，并完成自动分色。%A printed fabric automatic separation system based on SOM neural networks and im-proved K-means algorithm is proposed ,which can distinguish the color of color patterns on printed fabric to meet industry demand for fabric colors .In three-dimensional RGB color space , genetic algorithm is used for sub-images which represent the original fabric color distribution , thereby reducing the computation of color separation .Then two layers of clustering method is used to complete color separation :the first layer uses SOM neural network for a large number of data samples to complete the initial cluster ,which was much larger than expected number of clusters .The second layer uses the improved K-means algorithm to further clustering neurons , which are through the initial cluster of SOM neural network .Thus final clustering results are obtained and the separation is completed .Experimental results show that this system can accu-rately distinguish colors of pattern on printed fabric and complete automatic color separation .