Domain adaptation dark-light face detection algorithm based on improved SSD
Face detection techniques have been widely used in practical scenarios,however,there are many challenges in face detection under poor lighting conditions,including difficulty in extracting face features,poor discriminability,and unbalanced distribution of annotated data. Aiming at these problems,a domain adaptation dark-light face detection algorithm based on improved SSD is proposed. Firstly,image enhancement and domain adaptation method are used to reduce the domain difference between normal light data and dark-light data distributions. Secondly,an attention mechanism-based feature enhancement module is proposed based on the SSD detection framework. Finally,an asymptotic loss function is proposed for the original feature map and the enhanced feature map. Experimental results show that the method has a more competitive performance in the dark light environment compared with current mainstream face detection methods.
face detection algorithmdark facedomain adaptation modulefeature enhancement moduleattention mechanism
万方数据
维普
