Multi-task Face Attribute Recognition Based On Convolutional Neural Network

Face attribute recognition is an important research direction of face analysis task.The main goal is to analyze and predict the face attribute information in the face image.Because the appearance of human face will change dramatically with environment,emotion and other factors,face attribute recognition still faces many challenges.In recent years,the deep learning method based on convolutional neural network has become an important method of face attribute recognition with its powerful ability of image feature analysis.Due to the large number of human face attributes and the correlation between different attributes,the general deep learning method has great limitations,while the multi-task deep learning method is used for parallel prediction of multiple tasks,which conforms to the characteristics of human face attribute recognition,so it is of practical significance to study the multi-task human face attribute recognition method based on convolution neural network.The main contents of this paper are as follows:A multi task face attribute recognition method based on branch sharing is proposed.The traditional multi task neural network of branch structure ignores the high-level semantic information exchange between branches,and can not fully mine the relevance between related attribute tasks.Moreover,different nodes of the traditional branch structure have great differences,and the adaptability to different tasks is poor.In addition,due to the large number of face attributes,there must be an imbalance in the number of attribute samples.In order to solve the above problems,this paper proposes a branch sharing multi task face attribute recognition method,and verifies it on the celeba data set.The average recognition rate of face attribute is 91.9%.A multi task face recognition method based on joint assistant tasks is proposed.Face attribute recognition is one of the tasks of face analysis,and there are correlations among different tasks(face detection,face key point detection,border regression,face attribute recognition,etc.).In the traditional face attribute detection process,although some face analysis tasks are used to preprocess the image,the face attribute recognition tasks are independent of these face analysis tasks,which makes the model ignore the relevance between the face analysis tasks.Moreover,in the process of face attribute analysis,the background pixels of the image are easy to have ambiguous influence on the model.In order to solve the above problems,this paper proposes a multi task face attribute recognition method based on joint assistant tasks,and verifies it on celeba data set.The average recognition rate of attributes is 93.1%.