Owing to complex changes in face pose and the obvious influence on face recognition performance, a new approach is proposed for multi-pose face recognition based on the fusion of the LSTM (long short term memory network) and convolutional neural network-based cascade deep network (LCCDN) and incremental clustering. First, a LCCDN is designed to locate facial landmarks, and the memory function of the LSTM in LCCDN is used to explore the spatial contextual information between facial landmarks; then, facial landmarks are initialized. A CNN network model is used to fine facial landmarks by employing a coarse-to-fine strategy. Next, we consider the facial landmarks as face orientation descriptors. Simultaneously, to adapt to the dynamic updating of the face pose, an entropy-induced metric-based incremental clustering method is used to construct a face-pose pool by dynamically clustering head poses. In this manner, multi-pose face recognition is realized by establishing various face classification models with different poses. The recognition accuracies using the CAS-PEAL-R1, CFP, and Multi-PIE datasets are 96.75%, 96.50%, and 97.82%, respectively. In addition, comparisons with existing multi-pose face recognition methods highlight the superior performance of the proposed method.