Image Matching Based on Fast PCA-SIFT Descriptors with Automatic Determination of Dimensionality for PCA

Image matching is a key technology in the field of image processing. An effective image matching method based on fast PCA-SIFT descriptors is proposed and studied deeply. Firstly, the matrix left multiplication method is used to reduce the operation load of PCA and improve its operation speed. Secondly, we utilize the total interpretation proportion of the data variance of the top several principal components to determine the optimal dimension of the descriptor for PCA, thus the number of retained principal components can be determined automatically. Some intuitive and persuasive simulation experiments are carried out by using the proposed method. Experimental results demonstrate that the proposed method can automatically determine the number of retained principal components, and can reduce the operation load of PCA. The proposed image matching method can be used in the fields of three-dimensional reconstruction, cooperative augmented reality and teleoperation robots.