采用改进的小波和颜色矩的色纺面料图像检索

张霞 向军 张宁 潘如如

摘要： 传统的色纺面料检索方法效率低下、精度低，针对色纺面料颜色种类丰富、纹理复杂多变的特点，文章提出一种基于颜色矩和改进的小波变换结合的图像检索算法。采集色纺面料图像并建立标准图像数据库;分别采用颜色矩和改进的小波变换的方法提取色纺面料图像的颜色特征和纹理特征，使用欧氏距离测度的方法计算两种特征的子相似度;采用加权求和的方式获得图像的总体相似度，排序得到检索结果。试验中分别讨论了不同特征权重对检索结果的影响，验证了当颜色特征权重占0.8纹理特征权重占0.2时，检索结果最优。对比试验表明，该算法可兼顾颜色特征和纹理特征对检索结果的影响，平均检索精度mAP达到80.8%，优于现有方法。

关键词： 面料检索;特征提取;颜色矩;小波变换;色纺面料

中图分类号： TS941.26

文献标志码： A

文章编号： 1001-7003（2021）12-0034-06

引用页码： 121107

DOI： 10.3969/j.issn.1001-7003.2021.12.007（篇序）

Abstract： In view that traditional retrieval methods of colored spun fabrics are inefficient and inaccurate， while colored spun fabrics are characterized by rich colors and complex textures， an image retrieval algorithm combining color moment and modified wavelet transform was proposed. Firstly， colored spun fabric images were collected to establish a standard image database. Secondly， the color features and texture features of the colored spun fabric images were extracted using color moments and modified wavelet transform method respectively. The sub-similarity of color features and texture features were measured using Euclidean distance measurement method. Finally， the overall similarity of the images was obtained through weighted summation and the retrieval results were obtained by sorting.In the experiment， the influence of different feature weights on the retrieval results was discussed， and it was verified that when the color feature weights accounted for 0.8 and texture feature for 0.2 respectively， the retrieval result was optimal. The results of contrast experiment have shown that the algorithm can take into account the effects of color features and texture features on retrieval results， with an average retrieval precision ratio of 80.8% which is superior to the existing method.

Key words： fabric retrieval; feature extraction; color moment; wavelet transform; colored spun fabric

色纺面料是指由色纺纱织成的面料，具有独特的颜色和纹理效果，广泛应用于各类服饰、家纺产品。由于其丰富的颜色和纹理种类，纺织企业如何快速从中查找所需面料数据指导生产变得困难，图像检索是解决这一问题的有效方法。相比基于文本的图像检索，基于内容的图像检索是利用计算机视觉和图像处理技术提取图像的视觉特征（颜色、纹理及空间特征等）并映射为特征向量，根据特征向量之间的距离来衡量图像之间的相似度进行匹配[1]，避免了人工文本标注的主观性和工作量，检索结果更加准确和快速，能够及时准确获取已有产品的工艺单用于指导生产，缩短产品的生产周期，提高生产效率。……