Research On Robust Feature Extraction And Analysis Algorithms For Facial Attributes

Facial attribute recognition is using the feature which extracted from a given face image to predict the attribute information,such as facial expression,age,gender,race,etc.Based on the extracted features,it can auxiliary help machine to realize the identification or emotion analysis.The research on facial attribute recognition is one of the hotspots in the field of computer vision.With the development and application of the deep learning method,the research of facial attribute recognition also begins to use the deep learning method and obtains better recognition effect.Based on deep learning and making full use of local and global feature information of face images,this paper proposes three facial attribute feature extraction and analysis algorithms.The main work of this paper is as follows:(1)Proposed a Facial Attributes Analysis based on Deep Forest algorithm(FAA-DF).Deep forest algorithm has the effect of feature extraction,recognition and classification,and the algorithm has good generalization performance,which can be applied to different data sets,so it can be used for facial attribute robust feature extraction and analysis.In this paper,facial expression,age and gender feature extraction and analysis are carried out on CK+,RAF-DB,Adience and CelebA data sets,verifying the effectiveness of FAA-DF algorithm for facial attribute recognition.(2)Proposed a Facial Attributes Analysis based on Enhanced Deep Forest(FAA-EDF).The proposed algorithm improves its structure on the basis of deep forest,adds gradient boosting decision tree forest and logistic regression module,which can improve the performance of the algorithm,and has robustness for facial attributes.In this paper,facial expression,age and gender feature extraction and analysis are carried out on CK+,RAF-DB,Adience and CelebA data sets,and based on the properties of deep forest face feature extraction and analysis algorithms,FAA-EDF algorithm increased by 2.04%on the CK+data set,increased by 2.48%on the RAF-DB data set,on the Adience data set the age attributes,the property increased by 3.08%,gender attributes increased by 2.01%,on CelebA data set the age attributes,the property increased by 1.54%,the gender attributes increased by 1.84%,which verifies that the enhanced deep forest has better performance and stronger robustness for facial attribute recognition.(3)Proposed a Facial Attributes Analysis based on Heterogeneous Deep Model(FAA-HDM).The proposed heterogeneous deep model uses AlexNet deep neural network and enhanced deep forest to generate collaborative description fusion features by extracted deep neural network features and deep forest features respectively,which can be used for subsequent facial attribute recognition.In this paper,the robustness and effectiveness of the proposed FAA-HDM algorithm in facial attribute recognition are verified on CK+,RAF-DB,Adience and CelebA data sets.