Classroom face detection algorithm based on improved MTCNN

Aiming at the problem of poor face detection performance under the classroom scenario with different angle of views, occlusion, and uneven distribution of face scales, a novel classroom face detection method based on the improved multi-task cascaded convolutional neural network (MTCNN) algorithm is proposed in this paper. Firstly, a deep residual feature generation module is introduced to improve the detection accuracy of small-scale faces by utilizing the characteristics of low-level fine granularity and converting the original poor features into high-resolution deformation features. Then, all parts involving landmarks are removed to get the simplified MTCNN model, which is combined with deep residual feature generation module to improve the detection speed while the accuracy is ensured. Finally, an up-and-down cropping strategy is employed to solve the problem of large population and uneven face scale in the classroom scenario. Experimental results demonstrate that the proposed method can achieve superior accuracy and efficiency over some state-of-the-art approaches for face detection on the FDDB dataset, as well as in the real classroom scenario.