Research On Discrimination Image Recognition Method Based On Attribute Feature Fusion

Image is the carrier of rich information,which often reflects people's emotional content.Because of its complex causes,changeable scenes,various forms of expression and obvious differences among individuals,discrimination emotion recognition in images has become a difficult problem in the field of computer vision.In this paper,several problems existing in the task of discriminative emotion recognition in image are studied and improved,and a model of discrimination image recognition based on attribute feature fusion(DRRA+)is proposed,which includes the following contents:In view of the lack of standard discriminative emotion analysis data set,a discriminative image data set including dynamic and static attribute information of objects has been constructed.The data were collected by Internet and photographed independently.According to the static attributes of age,gender,skin color,facial expression and hand posture,2800 sets of discrimination data including age discrimination,gender discrimination,racial discrimination and non-discrimination were set up by manual annotation,and the discrimination emotion was divided into more finegrained ones.In view of the characteristics that the discriminative emotion of the characters in the discriminative image is often expressed by expression and gesture,a multi task relationship recognition model(DRR)based on dynamic attribute feature fusion is proposed.Based on the semantic transformation of the global character as the relationship feature,the model integrates the expression and gesture attribute information of the character to enhance the relationship representation.At the same time,the multi task loss module is added to make the model output the expression,gesture category and relationship category of characters at the same time,and the high-level semantic information of dynamic attributes of characters is used to assist the prediction of discrimination relationship.By comparing the social relation recognition(SRR),the validity of DRR model in image discrimination recognition task is proved.In view of the importance of age,gender and skin color attribute information in discriminative images to the fine-grained discrimination classification,this paper proposes a discrimination relationship recognition with attention(DRRA)model.In this model,attention mechanism is used to enhance the attention of key areas and channels in the image,and to improve the results of extraction and classification of age,gender and skin color attributes.Experiments show that DRRA method improves the effect of relationship prediction.Finally,the attention module and the dynamic attribute feature fusion module are combined to form the discrimination recognition model DRRA+.Through multiple groups of comparative experiments,it is proved that the combination of the two improved modules further improves the effectiveness of the algorithm model.