In this paper, a new technique named dynamic two-dimension principal component analysis (D2DPCA) is proposed for the Unmanned Underwater Vehicle (UUV's) underwater video image information. Compared with PCA and 2DPCA, not only considers the static relationship between image data, but also considers the dynamic characteristics between data are considered in 2DPCA. Thus, D2DPCA enables the information which has a short-time high-frequency sampling characteristic to be sufficiently expressed. Moreover, using image sample data with dynamic characteristics can directly construct the covariance matrix of the image and derive the projection vector of image feature extraction. In order to test D2DPCA and evaluate its performance, a series of experiments on image data are carried out by us. The experimental results show that D2DPCA has higher target recognition accuracy and more effective feature extraction capability than PCA and 2DPCA.
展开▼
