Compared with high-resolution （HR） face image， the low-resolution （LR） face image recognition effect is poorer. Researchers have put forward several LR face recognition algorithms based on the canonical correlation analysis （CCA） and kernel canonical correlation analysis （KCCA） to solve this problem， which ignored the supervised information and the consistency information between different views. In this paper， we put forward a novel dimensionality reduction algorithm—consistent discriminant correlation analysis （CDCA） by virtue of the class information and consistency information of different views. Furthermore， we design a LR face recognition algorithm based on CDCA. Concretely， we extract the principal component features from HR and LR face images respectively， use CDCA to learn the characteristic projection matrix of HR and LR face， and realize LR face recognition with the help of projection matrix. The experimental results show the superiority of the proposed method on recognition effect and robustness compared with the existing LR face recognition algorithms.