基于多子区策略的丝绸图像匹配算法

Silk image matching algorithm based on multiple sub-region strategy

印花是丝绸生产工艺中重要一环,基于图像的匹配技术可用于检验丝绸的印花质量。本文提出了一种基于多子区策略的丝绸彩色图像和矢量模板图匹配算法,并采用边缘信息作为丝绸彩色图像和矢量图的共性特征,解决了两种图像灰度不一致的问题。对预处理后的丝绸彩色图像提取图像边缘,选取边缘信息完整的区域作为待匹配子区,采用改进的Hausdorff距离矩阵作为最终的相似度准则。该算法可以获取较高的匹配正确率,基于真实丝绸场景数据的实验结果,证明了所提方法的有效性。

Silk textile industry is an important part of China’s light industry system,occupying a pivotal position in the national economy.The production and export scale of silk products is among the top in the global market.With the comprehensive promotion of"Made in China 2025",the combination of silk product manufacturing and industrialized intelligent detection has become the important means and development trend to improve product quality and reduce production costs.The matching method based on computer vision technology is an important means in the field of silk pattern printing.In the actual production environment,due to such factors as the position and angle of the light source,the flatness of the silk when placed and the shooting angle of the camera,the quality of the acquired silk images varies,thereby greatly influencing the final matching results.When handling silk patterns with complex backgrounds,the existing matching algorithm is not satisfactory in terms of the real-time and accuracy rate,making it unapplicable to the matching of silk images.There are three reasons.Firstly,the silk patterns are complicated and the backgrounds are complex.Secondly,it is difficult to directly apply pixel-based matching technique to the matching of template vector images,and thirdly,real-scenario silk pattern matching sets higher requirements for real-time and accuracy.In this paper,a matching algorithm for silk color images and vector template images based on multiple sub-region strategy was proposed,solving the grayscale inconsistency problem of the two images with edge information as the common feature of silk color images and vector images.Firstly,the angle of the image to be matched was adjusted by affine transformation to eliminate the angle difference with the vector template images to improve the matching efficiency.The edges of the pre-processed silk color images were extracted using Canny algorithm,the continuous edge density matrix of the edge pattern was calculated,the region with complete edge information was selected as the sub-region to be matched,and the geometric constraint relationship of the sub-region was calculated to improve the matching efficiency.The edge continuity of the image was commonly obscured,broken and unevenly illuminated in the process of silk image acquisition,affecting the matching results.Thus,we proposed a Hausdorff distance norm similarity criterion based on the continuous edge density matrix to describe the correlation degree of matching.Considering the great influence of edge information on the matching results,the edge continuity density of the image was used for the weighting of Hausdorff distance matrix,with the result being the final similarity measurement value.The obtained similarity degree of real-time silk images at different positions of the vector template images could be converted into the calculation of the Frobenius norm for the corresponding weighting matrix in order to conduct quantative comparison and finally the matching results are obtained.The algorithm can obtain a high matching accuracy.By testing the robustness and accuracy of the real-scenario silk patterns,the effectiveness of the proposed method is proved.