Hierarchical Discriminant Regression Tree Algorithm Based on BDPCA and its Application in Object Recognition
Weimin Ge, Kaikai Yuan, Xiaofeng Wang, Gang Wu · 2019
Aiming at the problem of the slow speed of the clustering and regression for high-dimensional data, the process of the hierarchical discriminant regression tree (HDR) is firstly analyzed. Then, the dimension reduction method of the traditional linear manifold was required to convert the image matrix into the high dimension vector, which greatly increases the processing time of samples. The bi-directional principal component analysis (BDPCA) algorithm can directly reduce the dimensions of the feature matrix and the processing time of samples. In the process of clustering in the output space, HDR tree randomly chooses the center points of input space, which leads to the depth of the HDR tree becoming larger and the balance of the tree becoming also worse. Therefore, we propose a new construction and retrieval algorithm, called 2-dimension HDR (2DHDR), which can reduce the depth of the tree and accelerate the processing speed of samples. Experiments on object block database and ORL face database show that the proposed 2DHDR has a faster construction and retrieval speed than the traditional HDR.