Research Of Face Parsing Based On Convolutional Neural Network

With the development of deep learning,various fields of computer vision have been greatly developed.Faces are more popular among scholars because they contain human biometrics and easy to collect.The application of face from the face detection,face recognition to the face interaction,face understanding,face parsing algorithm played a crucial role.The face parsing algorithm classifies each pixel of a single-frame image of a face,and achieves a more accurate effect than face detection.In order to solve the problem that the parsing effect of different face components in the face parsing task is poor,a face parsing algorithm based on edge fusion is proposed.By adding a new branch to the parsing model,the edge detection of the face component is performed to achieve the purpose of intensive training on the edge.At the same time,the edge information is combined with the parsing information to achieve better parsing results.In order to prove the effectiveness of the edge-fusion face parsing algorithm,the effect of not adding the edge information model is compared,and it is proved that the edge fusion has a significant effect on the model.By comparing the four segmentation algorithms in the academic world,the face parsing algorithm based on edge fusion achieves the optimal effect on the Helen face segmentation dataset.In order to reduce the time cost of face parsing dataset annotation,a face parsing dataset generation method is proposed.The method utilizes the location information of face landmarks to assist the generation of face parsing datasets.At the same time,in order to generate data more in line with face shape and structure,interpolation method and polynomial fitting method are used to optimize different parts of face.In order to prove the effectiveness of the method,the data generated by the face parsing dataset generation method and the data of the Helen dataset are compared.For the face hair region and the facial features,the data generated by the face parsing dataset generation method is obviously excellent than Helen dataset.