基于Mask R-CNN的人脸检测与分割方法

Face Detection and Segmentation Method Based on Mask R-CNN

针对现有主流的人脸检测算法不具备像素级分割,从而存在人脸特征具有噪声及检测精度不理想的问题,提出了一种基于Mask R-CNN的人脸检测及分割方法。通过ResNet-101结合RPN网络生成候选区域,再利用RoIAlign算法实现像素级的特征点定位,旨在提高定位精度。根据全卷积网络生成相应的人脸二值掩码,实现图像中人脸信息与背景的分割。此外,构建了一个具有分割标注信息的人脸数据集用于训练相应模型。在通用人脸检测数据集的实验结果表明,该方法具有较好的人脸检测效果,并能在准确检测的同时实现像素级的人脸信息分割。

Face detection is an important research direction in computer vision and information security,which has been widely studied over the past few decades.In the traditional face detection method,there is no pixel-level segmentation process,which leads to the problem of face features with noise and unsatisfactory detection accuracy.In order to overcome this shortcoming,a face detection and segmentation method based on Mask R-CNN is proposed in this paper.In this method,ResNet-101 and RPN is used to generate RoIs,and RoIAlign faithfully retains the exact spatial locations to generate binary mask through Fully Convolution Network.In order to train the model,this paper constructs a face dataset with segmentation annotation information.The experimental results of well-known face detection dataset show that the proposed method has better face detection effect and can achieve pixel-level face information segmentation at the same time.