基于多特征融合的织物印花图像分割  

Fabric image segmentation based on multi-feature fusion

作　　者：曹丽[1] 胡旭东[2] 

机构地区：[1]浙江理工大学信息学院,浙江杭州310018 [2]浙江理工大学机械与自动控制学院,浙江杭州310018

出　　处：《纺织学报》2016年第8期149-153,共5页Journal of Textile Research

摘　　要：针对织物印花检测精度的问题,采用结合颜色和纹理特征多特征融合的方法,对织物印花图像进行有效分割。在织物印花分割过程中,首先采用颜色特征结合基于自动种子点选取的区域增长算法对图像进行初始分割,在此基础上,利用小波变换提取干扰区域的纹理特征,从而可进一步消除干扰区域,实现织物印花图像的准确分割。实验结果表明:基于多特征融合的分割算法能够准确地分割出织物的印花图案,克服了仅仅采用颜色特征或者纹理特征时产生的分割失真,提高了分割质量,具有较好的应用价值。In order to improve the accuracy of fabric printing patterns,the paper studies an effective method based on the multi-feature fusion for printed fabric image segmentation.In the process of segmentation,an automatic seeded region growing algorithm combined with the color features are used to segment the image firstly.Due to the influence of disturbances,some printed regions may be lost by over segmentation in the image.After the initial segmentation,in order to improve the accuracy of segmentation,wavelet-based texture features are employed to retrieve the lost regions.The experimental results show that the proposed algorithm has good effect on the segmentation of printed fabric image,especially for the printed image having more textures and can eliminate the segmentation distortion caused by only using color feature or texture feature.Therefore,this algorithm has comparatively high practical value.

关 键 词：颜色特征 纹理特征 织物印花 图像分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]