Abstract: If a face image is interfered by redundant information, the accuracy of the extracted effective feature points from the image is not high enough. To solve this problem, a face feature point detection algorithm based on a cascade convolutional neural network is proposed in this paper. The algorithm reads in information of an original image via the input-function first, then extracts local features of the image through neurons in a receptive domain, and inputs all local features into a pooling domain. After doing that, the algorithm is able to average all captured local features stored in the pooling domain, and further do down-sampling to the pooling domain, and reduce the dimension of the convolution results, and finally output detection results of those feature points by virtual of iterative training. In this study, Python language is used to program the algorithm, and simulation experiments are carried out with the help of face data set. Simulation results show that the algorithm has a high recognition rate for face feature points.

Key words: convolutional neural network, face feature point detection, image recognition, convolutional domain