Face Detection Algorithm Based On Image Enhancement

Face detection is the first step of face recognition and other related applications,but the performance of most of the algorithm in low light image is seriously reduced,so improving the performance of face detection algorithm in low light image is particularly important for new applications such as automatic driving.In this paper,a cascade algorithm of enhancement and detection is proposed to improve the performance of low light face detection algorithm.Firstly,a low light face detection algorithm based on Retinex model is designed.The algorithm is divided into two steps.The first step is to process the low light image based on Retinex model,and explore the effect of single scale and multi-scale Retinex model on the enhancement of low light image.The second step is to use the enhanced image for face detection.In the face detection part,the improved SSD object detection algorithm is used.Aiming at the problem of sample imbalance,anchor refinement module is used to filter a large number of simple negative samples to reduce the search space,Then the location of the detection frame is further refined.The performance of low light face detection is improved through the combination of image enhancement and face detection.Secondly,a low light face detection algorithm based on the generative adversarial network is designed.The low light image is transformed into the normal light image by a generator with a skip-connection layer encoder decoder structure,and then the generated image is detected.In order to ensure the effect of generative network low light image enhancement,we use global and local discriminators to ensure that the enhancement effects of local and global image are consistent,to improve the image detail processing of the model,and use relative discriminator and perceptual loss to optimize the image quality.Finally,we use AP to measure the performance of low light face detection algorithm.The experimental results show that the low light image enhancement algorithm based on Retinex model can improve the performance AP of face detection from 16.43%to 19.67%,with an increase of 19.72%;the low light image enhancement algorithm based on the generative adversarial network can improve the performance AP of face detection from 16.43%to 20.41%,with an increase of 24.22%.The experimental results show that the proposed algorithm is effective in improving the performance of low light face detection.