A Multi-task Cascaded Algorithm with Optimized Convolution Neural Network for Face Detection

In this paper, a face detection algorithm based on optimizing Multi-task cascaded convolution neural network (MTCNN) was proposed. In particular, be based on the CNN structure of MTCNN, using its optimized cascaded CNN to form the framework of this. The two stages before are used to predict face, and the third network needs to do face prediction and landmark location. Besides, Mosaic technology had being used to adjust the data distribution of the dataset. Our method achieves accuracy over the original techniques on the challenging FDDB by 2.7%, reaching 97.6%, while our model size is reduced by about 1%.