In recent years, cat face recognition has received more and more attention. As an effective dimensional reduction method, manifold learning can help us better understand and deal with high-dimensional data. In this paper, unsupervised manifold learning based on polynomial projection is introduced for cat face recognition. Firstly, the data matrices are acquired by transforming the processed cat face picture into a matrix. Through the selection of datum points, the information coordinates are obtained which construct the polynomial kernel matrix. Next, after the polynomial projection manifold learning algorithm for cat face is proposed, it is applied to get the low-dimensional representation. Finally, carry out cat face recognition on the Graphical User Interface(GUI) and the recognition accuracy is obtained by computation simulations. Compared with the traditional dimensional reduction method, the presented method can provide a more accurate means to achieve cat face recognition which accomplishes the purpose of helping stray cats to go back home.