Face detection algorithm under low-light based on feature recovery

Face detection detects and locates faces in images for face recognition, face tracking, and analysis applications. The performance of many advanced face recognition models deteriorates significantly when applied to low-light environments, hence face detection from low-light images is challenging. To solve the problem, this paper proposes a face detection method based on feature recovery, which includes two modules: feature recovery and feature extraction. The feature recovery module can obtain the face feature recovery image, which is fused with the original low-light face image to obtain the face feature image. On this basis, the feature extraction is trained for face detection. Finally, a face detection method suitable for low-light is obtained. It solves the difficulty of face detection under low-light. The experiment results carried out the overall detection precision increased by 18% on the DARK FACE test set, which verified the effectiveness of the proposed method.