Cloth-independent features are the main core of Cloth-Changing Person Re-Identification (CCreID), such as the face, hairstyle, gaits, body structure, shape, etc. Most current CCreID models exploit multi-modality information (e.g., contour, skeleton, and silhouette) to estimate the body shape for target matching. However, the performance and complexity of these methods are highly dependent on the additional multi-modality models (e.g., accuracy, and computational cost). This paper explores cloth-independent features directly from the original images through a multi-perspective of identity analysis. To this end, we propose a novel Cloth-Independent feature learning from Multi-Perspective for Cloth-Changing Person Re-Identification (CIMP-CCreID) model. Specifically, we introduce View Classification Learning (VCL) and Cloth Classification Learning (CCL) along with identity classification, which force the model to mine cloth-independent features from different perspectives of identity in RGB images. Since no view ground truths are available in the current cloth-changing datasets, we propose a new View Predictor (VP) module that predicts the identity view in the image with respect to the capturing camera. Extensive experiments on three benchmarked cloth-changing datasets, including LTCC, PRCC, and VC-Clothes, demonstrate the effectiveness of our model against state-of-the-art CCreID methods.