Abstract:

In order to solve the problems of low matching efficiency, high time complexity and high computational complexity in image matching algorithms, an image matching algorithm is proposed by combining sparse representation and topological similarity. The algorithm first performs feature detection on the image, calculates the contour similarity, finds the largest contour area similar in the image to be matched, uses sparse coding to sparsely represent the features in the contour, establishes a sparse model, and simplifies complex features, but without affecting the classification method of features, the features of the same category or the same attribute are classified into the same feature set, and the sparse representation and the mutual attribute learning of the neighboring mutual information are combined to learn. The transformation matrix is calculated to represent the image, and the structured topological similarity is used to optimize the points associated with the inside and outside of the contour. Finally, the algorithm is analyzed from subjective evaluation and objective evaluation. The results show that compared with other image matching algorithms, the proposed new algorithm has obvious matching accuracy and effect. It is concluded that the new algorithm has better advantages in improving matching efficiency and complexity.

Key words: contour similarity, sparse representation, topological similarity, image matching

摘要：

为了解决图像匹配算法中存在的匹配效率低、时间复杂度与计算量高等问题，通过结合稀疏表示和拓扑相似性，提出了一种图像匹配算法。该算法先对图像进行特征检测，计算轮廓相似度，找到待匹配图像中相似的最大轮廓区域，用稀疏编码对轮廓内特征进行稀疏表示，建立稀疏模型，将复杂特征变得单一化，但又不影响特征的分类方式，将相同类别或者相同属性的特征归为同一特征集，结合稀疏表示和邻域互信息的类属属性学习。计算得到变换矩阵，用以表示图像。利用结构化的拓扑相似性，对轮廓内外相关联的点进行优化。最后，分别从主观评价和客观评价两个方面对算法进行分析，结果表明提出的新算法与其他图像匹配算法相比较，具有明显匹配精度与效果，提出的算法在提高匹配效率及复杂度等方面具有较好优势。

关键词: 轮廓相似度, 稀疏表示, 拓扑相似性, 图像匹配