基于交互式分割的蕾丝花型检索算法

Lace retrieval algorithm based on interactive segmentation

由于蕾丝花边工艺复杂,花型图案种类繁多,且传统的基于文本的图像检索方法采用人工标注,工作量大、主观性强,不能满足蕾丝花型的检索需求,因此文章采用基于Live Wire交互式图像分割和层次匹配的方法对蕾丝花型进行检索。首先利用Live Wire算法提取蕾丝花边中一个或多个花型轮廓;然后分别提取傅立叶描述子、几何特征、灰度共生矩阵作为其特征向量;最后将每个测试花型的特征向量与1 327个样本花型的特征向量进行匹配并按相似度大小排序。实验结果表明,相比于单一的特征识别,该方法具有更高的识别率,能够满足蕾丝花型的检索需求。

The lace process is complex,and pattern types are numerous.Besides,traditional text-based image retrievalis manually annotated,which has the advantages of heavy workload and strong subjectivity.So,it cannot meet the retrieval demand of lace.Therefore,a novel method of retrieving lace patterns based on Live Wire interactive image segmentation and hierarchical matching is proposed in this paper.One or more pattern contours were extracted by using Live Wire algorithms firstly.Then,Fourier descriptors,geometric features and gray level co-occurrence matrix were extracted respectively as feature vectors.Lastly,feature vectors of each testing pattern were matched with those of 1327 sample patterns according to similarity degree.The results show that the method proposed here presents much higher recognition rate compared with single characteristic recognition,which could meet higher standards of lace pattern retrieval.