基于高阶统计量的DAISY特征向量降维

Dimensionality Reduction of DAISY Feature Vectors based on High-Order Statistics

采用DAISY算子描述特征点时,每个特征点会生成一个1×200维度的特征向量。维度较高的特征向量会对后续的工作如特征点匹配等,带来非常大的计算量,严重影响算法的效率。因此,需要采取一定的方法降低特征向量的维度。因此,提出了一种基于三阶统计量的方法。这种方法可以通过提取原始向量中的主成分来降低维度。数值实验中证明,相对于经典的PCA降维算法,所提算法在提取主成分方面有更好的效果,同时可将向量的维数降到更低水平,大大提高了算法效率。

When the DAISY operator is used to describe the feature points, each feature point would generate a feature vector of 1x200 dimensionality. The feature vectors with higher dimensions would usually bring great computation to the following tasks, such as feature point matching, etc., and seriously affect the efficiency of the algorithm. It is necessary to adopt some methods and reduce the dimensionality of feature vectors. And for this reason, the method based on three-order statistics is proposed. This method can reduce the dimensionality by extracting the principal components in the original vector. Numerical experiments indicate that compared with the classic PCA dimensionality reduction algorithm, the proposed algorithm has better effect in extracting the principal components, and could reduce the dimensionality of vectors to a lower level, thus greatly improving the efficiency of this algorithm.