基于大数据聚类的化妆品包装符号元素特征提取

Feature extraction of cosmetic packaging symbol elements based on big data clustering

研究了基于大数据聚类的化妆品包装符号元素特征提取方法。初始化处理水平集函数,获取化妆品包装符号元素图像的局部标准差图像,根据余弦相似性确定局部标准差图像像素点在轮廓曲线上的演化方向,水平集演化改进SPF函数,完成符号元素分割;计算大数据的离散样本频谱特征,由置信度获取数据聚类中心的粒子最优解的向量矩阵,完成数据聚类;结合卷积神经网络和AutoEncoder,通过卷积、过滤以及池化操作,在输出层存在的节点中获取最大激活值,实现化妆品包装符号元素特征的提取。实验结果表明,所提方法的特征提取时间较短、特征辨识力有所提高且提取准确率较高。

The feature extraction method of cosmetic packaging symbol elements based on big data clustering was proposed.The level set function was initialized to obtain the local standard deviation image of cosmetic packaging symbol element image.The evolution direction of local standard deviation image pixels on contour curve was determined according to the cosine similarity.The level set evolution improved the SPF function to complete the symbol element segmentation.The discrete sample spectrum characteristics of big data were calculated,and the vector matrix of the particle optimal solution of the data clustering center was obtained by the confidence degree to complete the data clustering.By combining convolution neural network and AutoEncoder,the maximum activation value was obtained in the nodes existing in the output layer through convolution,filtering and pooling operations,and the feature extraction of cosmetic packaging symbol elements was realized.The experimental results show that the proposed method has shorter feature extraction time,higher feature recognition ability and higher extraction accuracy.