韩锐.基于不变矩特征模型耦合相似度量规则的图像匹配算法[J].包装工程,2018,39(9):204-211.
HAN Rui.An Image Matching Algorithm Based on Invariant Moment Feature Model and Coupling Similarity Measurement Rule[J].Packaging Engineering,2018,39(9):204-211. |
| 基于不变矩特征模型耦合相似度量规则的图像匹配算法 |
| An Image Matching Algorithm Based on Invariant Moment Feature Model and Coupling Similarity Measurement Rule |
| 投稿时间：2017-11-02 修订日期：2018-05-10 |
| DOI：10.19554/j.cnki.1001-3563.2018.09.035 |
| 中文关键词: 图像匹配 检测规则 不变矩特征模型 相似度量规则 RANSAC算法 Euclidean模型 |
| 英文关键词: image matching detection rule invariant moment feature model similarity measurement rule RANSAC algorithm Euclidean model |
| 基金项目:江苏省自然科学基金（BK20151102A）；江苏省产学研前瞻性联合研究项目（BY2013063-02） |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 为了解决当前图像匹配算法因主要利用特征点之间的距离来实现特征匹配，从而忽略了特征点的结构特征，导致算法存在较多的漏匹配点以及错误匹配点等不足的问题。方法 提出基于不变矩特征模型耦合相似度量规则的图像匹配算法。通过对待检测像素点构造的邻域圆上的点进行分类，制定检测规则，对FAST算子进行改进，利用改进的FAST算子快速、精准地检测图像的特征点。随后，构造不变矩特征模型，取代SIFT算法中获取特征向量的方法，生成低维度的特征描述符。通过Euclidean模型和SSIM建立相似度量规则，对特征点之间的相似度进行度量，完成图像的特征匹配。最后，引入随机抽样一致性（RANSAC）算法剔除错误匹配点，完成图像的匹配。结果 仿真结果显示，相较于当前的图像匹配算法，所提算法具有更高的匹配正确度和鲁棒性，其查全率最高可达95%左右，且匹配效率较快，约为3.75 s。结论 所提匹配方法具备良好的匹配精度，在图像信息安全、包装条码识别与拼接等领域具有一定的参考价值。 |
| 英文摘要: |
| The work aims to solve the problems of many missing matching points and wrong matching points of the algorithm because the current image matching algorithm mainly achieves the feature matching by the distance between the feature points, thus ignoring the structural features of the feature point. An image matching algorithm based on invariant moment feature model and coupling similarity measurement rules was proposed. The detection rules were formulated by the classification of neighborhood circle points constructed by the pixel points to be detected, and the FAST descriptor was improved. The improved FAST descriptor was used to detect the feature points of the image quickly and precisely. Subsequently, the invariant moment feature model was constructed to replace the method for obtaining the feature vectors from SIFT algorithm to generate feature descriptors of lower dimension. Euclidean model and SSIM model were used to establish similarity measurement rules for the feature matching of images. Finally, the Random Sample Consensus (RANSAC) algorithm was introduced to eliminate the wrong matching points and complete the image matching. The simulation results showed that, compared with the current image matching algorithm, the proposed algorithm had higher matching accuracy and stronger robustness. Its maximum recall ratio could be around 95%, and it had faster matching efficiency, approximately 3.75 s. With good matching accuracy, the proposed matching method has certain reference value in the fields of image information security, packaging barcode recognition and splicing, etc. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
渝公网安备 50010702501716号