Research On Aesthetic Quality Evaluation Of Group Photo Image

Aesthetic quality evaluation of images has been a hot research topic in recent years.Although there are many research findings,most of the existing methods of aesthetic quality evaluation of images only focus on the general aesthetic characteristics of images,and do not achieve the desired effect on the aesthetic evaluation of group photo images in this specific field.However,group photo shooting has become a common practice in people’s daily communication and life.How to evaluate the aesthetic quality of a group photo image will become very meaningful.The existing methods of aesthetic evaluation of group photos only integrate aesthetic properties and group photo characteristics,and do not consider different aesthetic evaluation of group photos with different group photo shapes.In order to better evaluate the aesthetics of group photo images,this paper first classifies group photo images into five types of shapes according to the shape of group photo,which are “monogram type”,“round type”,“back type”,“jumping type” and “looking up type”.Influenced by a priori knowledge,we found that the head and pose can be used to distinguish the group photo shapes,and designed a deep learning method to classify the shapes.The model is constructed by borrowing the head detection model for target detection and the model of human pose recognition network.For this purpose,we constructed a group photo styling dataset with about 998 images,including domestic and international image libraries,existing aesthetic datasets(AVA,AADB,etc.)and the group’s own photography.The experimental results show that the extraction of head features and pose features can classify different group photo images well,and the accuracy of group photo styling classification reaches 93.9%,which is much better than the previous classification models.Second,different aesthetic evaluation methods were determined for each shape.To this end,we constructed a dataset for aesthetic evaluation of ensemble styling,and performed aesthetic evaluation of ensemble styling on the original dataset.In image aesthetic quality evaluation,the aesthetic features of concern are different for different scenes.Inspired by the aesthetic evaluation of multiple scenes,a group photo styling recognition network is introduced for the type determination of group photo styling.Since the group photo in this particular scene,not only the aesthetic properties of the background are considered,but also the face and posture of the characters need to be paid attention to.For this reason,we design a learning network that incorporates group photo styling recognition network and group photo manual features,and deep aesthetic learning features.The model is composed of manual features borrowed from Baidu AI and Face++,deep learning features from NIMA model with the model of group photo shape recognition.The experimental results show that after the group photo modeling classification,it is possible to perform a good categorical aesthetic evaluation of different group photo images;by comparing different aesthetic evaluation methods,it is also proved that the categorical evaluation can evaluate the group photo images better than taking a single evaluation method.This enriches the application field of aesthetic evaluation.